# Eight to Late

Sensemaking and Analytics for Organizations

## A gentle introduction to random forests using R

with 7 comments

### Introduction

In a previous post, I described how decision tree algorithms work and demonstrated their use via the rpart library in R. Decision trees work by splitting a dataset recursively. That is, subsets arising from a split are further split until a predetermined termination criterion is reached.  At each step, a split is made based on the independent variable that results in the largest possible reduction in heterogeneity of the dependent  variable.

(Note:  readers unfamiliar with decision trees may want to read that post before proceeding)

The main drawback of decision trees is that they are prone to overfitting.   The  reason for this is that trees, if grown deep, are able to fit  all kinds of variations in the data, including noise. Although it is possible to address this partially by pruning, the result often remains less than satisfactory. This is because the algorithm makes a locally optimal choice at each split without any regard to whether the choice made is the best one overall.  A poor split made in the initial stages can thus doom the model, a problem that cannot be fixed by post-hoc pruning.

In this post I describe random forests, a tree-based algorithm that addresses the above shortcoming of decision trees. I’ll first describe the intuition behind the algorithm via an analogy and then do a demo using the R randomForest library.

### Motivating random forests

One of the reasons for the popularity of decision trees is that they reflect the way humans make decisions: by weighing up options at each stage and choosing the best one available.  The analogy is particularly useful because it also suggests how decision trees can be improved.

One of the lifelines in the game show, Who Wants to be A Millionaire, is “Ask The Audience” wherein a contestant can ask the audience to vote on the answer to a question.  The rationale here is that the majority response from a large number of independent decision makers is more likely to yield a correct answer than one from a randomly chosen person.  There are two factors at play here:

1. People have different experiences and will therefore draw upon different “data” to answer the question.
2. People have different knowledge bases and preferences and will therefore draw upon different “variables” to make their choices at each stage in their decision process.

Taking a cue from the above, it seems reasonable to build many decision trees using:

1. Different sets of training data.
2. Randomly selected subsets of variables at each split of every decision tree.

Predictions can then made by taking the majority vote over all trees (for classification problems) or averaging results over all trees (for regression problems).  This is essentially how the random forest algorithm works.

The net effect of the two strategies is to reduce overfitting by a) averaging over trees created from different samples of the dataset and b) decreasing the likelihood of a small set of strong predictors dominating the splits.  The price paid is reduced interpretability as well as increased computational complexity. But then, there is no such thing as a free lunch.

### The mechanics of the algorithm

Although we will not delve into the mathematical details of the algorithm, it is important to understand how two points made above are implemented in the algorithm.

#### Bootstrap aggregating… and a (rather cool) error estimate

A key feature of the algorithm is the use of multiple datasets for training individual decision trees.  This is done via a neat statistical trick called bootstrap aggregating (also called bagging).

Here’s how bagging works:

Assume you have a dataset of size N.  From this you create a sample (i.e. a subset) of size n (n less than or equal to N) by choosing n data points randomly with replacement.  “Randomly” means every point in the dataset is equally likely to be chosen and   “with replacement” means that a specific data point can appear more than once in the subset. Do this M times to create M equally-sized samples of size n each.  It can be shown that this procedure, which statisticians call bootstrapping, is legit when samples are created from large datasets – that is, when N is large.

Because a bagged sample is created by selection with replacement, there will generally be some points that are not selected.  In fact, it can be shown that, on the average, each sample will use about two-thirds of the available data points. This gives us a clever way to estimate the error as part of the process of model building.

Here’s how:

For every data point, obtain predictions for trees in which the point was out of bag. From the result mentioned above, this will yield approximately M/3 predictions per data point (because a third of the data points are out of bag).  Take the majority vote of these M/3 predictions as the predicted value for the data point. One can do this for the entire dataset. From these out of bag predictions for the whole dataset, we can estimate the overall error by computing a classification error (Count of correct predictions divided by N) for classification problems or the root mean squared error for regression problems.  This means there is no need to have a separate test data set, which is kind of cool.  However, if you have enough data, it is worth holding out some data for use as an independent test set. This is what we’ll do in the demo later.

#### Using subsets of predictor variables

Although bagging reduces overfitting somewhat, it does not address the issue completely. The reason is that in most datasets a small number of predictors tend to dominate the others.  These predictors tend to be selected in early splits and thus influence the shapes and sizes of a significant fraction of trees in the forest.  That is, strong predictors enhance correlations between trees which tends to come in the way of variance reduction.

A simple way to get around this problem is to use a random subset of variables at each split. This avoids over-representation of dominant variables and thus creates a more diverse forest. This is precisely what the random forest algorithm does.

### Random forests in R

In what follows, I use the famous Glass dataset from the mlbench library.  The dataset has 214 data points of six types of glass  with varying metal oxide content and refractive indexes. I’ll first build a decision tree model based on the data using the rpart library (recursive partitioning) that I covered in an earlier article and then use then show how one can build a random forest model using the randomForest library. The rationale behind this is to compare the two models – single decision tree vs random forest. In the interests of space,  I won’t explain details of the rpart here as  I’ve covered it at length in the previous article. However, for completeness, I’ll list the demo code for it before getting into random forests.

#### Decision trees using rpart

Here’s the code listing for building a decision tree using rpart on the Glass dataset (please see my previous article for a full explanation of each step). Note that I have not used pruning as there is little benefit to be gained from it (Exercise for the reader: try this for yourself!).

#set working directory if needed (modify path as needed)
setwd(“C:/Users/Kailash/Documents/rf”)
#load required libraries – rpart for classification and regression trees
library(rpart)
#mlbench for Glass dataset
library(mlbench)
#load Glass
data(“Glass”)
#set seed to ensure reproducible results
set.seed(42)
#split into training and test sets
Glass[,”train”] <- ifelse(runif(nrow(Glass))<0.8,1,0)
#separate training and test sets
trainGlass <- Glass[Glass$train==1,] testGlass <- Glass[Glass$train==0,]
#get column index of train flag
trainColNum <- grep(“train”,names(trainGlass))
#remove train flag column from train and test sets
trainGlass <- trainGlass[,-trainColNum]
testGlass <- testGlass[,-trainColNum]
#get column index of predicted variable in dataset
typeColNum <- grep(“Type”,names(Glass))
#build model
rpart_model <- rpart(Type ~.,data = trainGlass, method=”class”)
#plot tree
plot(rpart_model);text(rpart_model)
#…and the moment of reckoning
rpart_predict <- predict(rpart_model,testGlass[,-typeColNum],type=”class”)
mean(rpart_predict==testGlass$Type) [1] 0.6744186 Now, we know that decision tree algorithms tend to display high variance so the hit rate from any one tree is likely to be misleading. To address this we’ll generate a bunch of trees using different training sets (via random sampling) and calculate an average hit rate and spread (or standard deviation). #function to do multiple runs multiple_runs <- function(train_fraction,n,dataset){ fraction_correct <- rep(NA,n) set.seed(42) for (i in 1:n){ dataset[,”train”] <- ifelse(runif(nrow(dataset))<0.8,1,0) trainColNum <- grep(“train”,names(dataset)) typeColNum <- grep(“Type”,names(dataset)) trainset <- dataset[dataset$train==1,-trainColNum]
testset <- dataset[dataset$train==0,-trainColNum] rpart_model <- rpart(Type~.,data = trainset, method=”class”) rpart_test_predict <- predict(rpart_model,testset[,-typeColNum],type=”class”) fraction_correct[i] <- mean(rpart_test_predict==testset$Type)
}
return(fraction_correct)
}
#50 runs, no pruning
n_runs <- multiple_runs(0.8,50,Glass)
mean(n_runs)
[1] 0.6874315
sd(n_runs)
[1] 0.0530809

The decision tree algorithm gets it right about 69% of the time with a variation of about 5%. The variation isn’t too bad here, but the accuracy has hardly improved at all (Exercise for the reader: why?). Let’s see if we can do better using random forests.

#### Random forests

As discussed earlier, a random forest algorithm works by averaging over multiple trees using bootstrapped samples. Also, it reduces the correlation between trees by splitting on a random subset of predictors at each node in tree construction. The key parameters for randomForest algorithm are the number of trees (ntree) and the number of variables to be considered for splitting (mtry).  The algorithm sets a default of 500 for ntree and sets mtry to the  square root of the the number of predictors for classification problems or one-third the total number of predictors for regression.   These defaults can be overridden by explicitly providing values for these variables.

The preliminary stuff – the creation of training and test datasets etc. – is much the same as for decision trees but I’ll list the code for completeness.

#load required library – randomForest
library(randomForest)
#mlbench for Glass dataset – load if not already loaded
#library(mlbench)
#load Glass
data(“Glass”)
#set seed to ensure reproducible results
set.seed(42)
#split into training and test sets
Glass[,”train”] <- ifelse(runif(nrow(Glass))<0.8,1,0)
#separate training and test sets
trainGlass <- Glass[Glass$train==1,] testGlass <- Glass[Glass$train==0,]
#get column index of train flag
trainColNum <- grep(“train”,names(trainGlass))
#remove train flag column from train and test sets
trainGlass <- trainGlass[,-trainColNum]
testGlass <- testGlass[,-trainColNum]
#get column index of predicted variable in dataset
typeColNum <- grep(“Type”,names(Glass))
#build model
Glass.rf <- randomForest(Type ~.,data = trainGlass, importance=TRUE, xtest=testGlass[,-typeColNum],ntree=1000)
#Get summary info
Glass.rf
Call:
randomForest(formula = Type ~ ., data = trainGlass, importance = TRUE, xtest = testGlass[, -typeColNum], ntree = 1001)
Type of random forest: classification
Number of trees: 1000
No. of variables tried at each split: 3
OOB estimate of error rate: 23.98%
Confusion matrix:
 1 2 3 5 6 7 class.error 1 40 7 2 0 0 0 0.1836735 2 8 49 1 2 2 1 0.2222222 3 6 3 6 0 0 0 0.6000000 5 0 1 0 11 0 1 0.1538462 6 1 2 0 1 6 0 0.5000000 7 1 2 0 1 0 21 0.1600000

The first thing to note is the out of bag error estimate is ~ 24%.  Equivalently the hit rate is 76%, which is better than the 69% for decision trees. Secondly, you’ll note that the algorithm does a terrible job identifying type 3 and 6 glasses correctly. This could possibly be improved by a technique called boosting, which works by  iteratively improving poor predictions made in earlier stages. I plan to look at boosting in a future post, but if you’re curious, check out the gbm package in R.

Finally, for completeness, let’s see how the test set does:

#accuracy for test set
mean(Glass.rf$test$predicted==testGlass$Type) [1] 0.8372093 #confusion matrix table(Glass.rf$test$predicted,testGlass$Type)
 1 2 3 5 6 7 1 19 2 0 0 0 0 2 1 9 1 0 0 0 3 1 1 1 0 0 0 5 0 1 0 0 0 0 6 0 0 0 0 3 0 7 0 0 0 0 0 4

The test accuracy is better than the out of bag accuracy and there are some differences in the class errors as well. However, overall the two compare quite well and are significantly better than the results of the decision tree algorithm.

### Variable importance

Random forest algorithms also give measures of variable importance. Computation of these is enabled by setting  importance, a boolean parameter, to TRUE. The algorithm computes two measures of variable importance: mean decrease in Gini and mean decrease in accuracy. Brief explanations of these follow.

#### Mean decrease in Gini

When determining splits in individual trees, the algorithm looks for the largest class (in terms of population) and attempts to isolate it first. If this is not possible, it tries to do the best it can, always focusing on isolating the largest remaining class in every split.This is called the Gini splitting rule (see this article for a good explanation of the rule).

The “goodness of split” is measured by the Gini Impurity, $I_{G}$. For a set containing K categories this is given by:

$I_{G} = \sum_{i=1}^{K} f_{i}(1-f_{i})$

where $f_{i}$ is the fraction of the set that belongs to the ith category. Clearly, $I_{G}$  is 0 when the set is homogeneous or pure (1 class only) and is maximum when classes are equiprobable (for example, in a two class set the maximum occurs when $f_{1}$ and $f_{2}$ are 0.5). At each stage the algorithm chooses to split on the predictor that leads to the largest decrease in $I_{G}$. The algorithm tracks this decrease for each predictor for all splits and all trees in the forest. The average is reported  as the mean decrease in Gini.

#### Mean decrease in accuracy

The mean decrease in accuracy is calculated using the out of bag data points for each tree. The procedure goes as follows: when a particular tree is grown, the out of bag points are passed down the tree and the prediction accuracy (based on all out of bag points) recorded . The predictors are then randomly permuted and the out of bag prediction accuracy recalculated. The decrease in accuracy for a given predictor is the difference between the accuracy of the original (unpermuted) tree and the those obtained from the permuted trees in which the predictor was excluded. As in the previous case, the decrease in accuracy for each predictor can be computed and tracked as the algorithm progresses.  These can then be averaged by predictor to yield a mean decrease in accuracy.

#### Variable importance plot

From the above, it would seem that the mean decrease in accuracy is a more global measure as it uses fully constructed trees in contrast to the Gini measure which is based on individual splits. In practice, however, there could be other reasons for choosing one over the other…but that is neither here nor there, if you set importance to TRUE, you’ll get both. The numerical measures of importance are returned in the randomForest object (Glass.rf in our case), but I won’t list them here. Instead, I’ll just print out the variable importance plots for the two measures as these give a good visual overview of the relative importance of variables. The code is a simple one-liner:

#variable importance plot
varImpPlot(Glass.rf)

The plot is shown in Figure 1 below.

Figure 1: Variable importance plots

In this case the two measures are pretty consistent so it doesn’t really matter which one you choose.

### Wrapping up

Random forests are an example of a general class of techniques called ensemble methods. These techniques are based on the principle that averaging over a large number of not-so-good models  yields a more reliable prediction than a single model. This is true only if models in the group are independent of  each other, which is precisely what bootstrap aggregation and predictor subsetting are intended to achieve.

Although  considerably more complex than decision trees, the logic behind random forests is not hard to understand. Indeed, the intuitiveness of the algorithm together with its ease of use and accuracy have made it very popular in the machine learning community.

Written by K

September 20, 2016 at 9:44 pm

## The story before the story – a data science fable

with 5 comments

It is well-known that data-driven stories are a great way to convey results of data science initiatives. What is perhaps not as well-known is that data science projects often have to begin with stories too. Without this “story before the story” there will be no project, no results and no data-driven stories to tell….

For those who prefer to read, here’s a transcript of the video in full:

In the beginning there is no data, let alone results…but there are ideas. So, long before we tell stories about data or results, we have to tell stories about our ideas. The aim of these stories is to get people to care about our ideas as much as we do and, more important, invest in them. Without their interest or investment there will be no results and no further stories to tell.

So one of the first things one has to do is craft a story about the idea…or the story before the story.

Once upon a time there was a CRM system. The system captured every customer interaction that occurred, whether it was by phone, email or face to face conversation. Many quantitative details of interactions were recorded, time, duration, type. And if the interaction led to a sale, the details of the sale were recorded too.

Almost as an aside, the system also gave sales people the opportunity to record their qualitative impressions as free text notes. As you might imagine, this information, though potentially valuable, was never analysed. Sure managers looked at notes in isolation from time to time when referring.to specific customer interactions, but there was no systematic analysis of the corpus as a whole. Nobody had thought it worthwhile to do this, possibly because it is difficult if not quite impossible to analyse unstructured information in the world of relational databases and SQL.

One day, an analyst was browsing data randomly in the system, as good analysts sometimes do. He came across a note that to him seemed like the epitome of a good note…it described what the interaction was about, the customer’s reactions and potential next steps all in a logical fashion.

This gave him an idea. Wouldn’t it be cool, he thought, if we could measure the quality of notes? Not only would this tell us something about the customer and the interaction, it may tell us something about the sales person as well.

The analyst was mega excited…but he realised he’d need help. He was an IT guy and as we all know, business folks in big corporations stopped listening to their IT guys long ago. So our IT guy had his work cut out for him.

After much cogitation, he decided to enlist the help of his friend, a strategic business analyst in the marketing department. This lady, who worked in marketing had the trust of the head of marketing. If she liked the idea, she might be able to help sell it to the head of marketing.

As it turned out, the business analyst loved the idea…more important, since she knew what the sales people do on a day to day basis, she could give the IT guy more ideas on how he could build quantitative measures of the quality of notes. For example, she suggested looking for  emotion-laden words or mentions of competitor’s products and so on. The IT guy now had some concrete things to work on.  The initial results gave them even more ideas, and soon they had more than enough to make a convincing pitch to the head of marketing.

It would take us too far afield to discuss details of the pitch, but what we will say is this: they avoided technical details, instead focusing on the strategic and innovative aspects of the work.

The marketing head liked the idea…what was there not to like? He agreed to support the effort, and the idea became a project….

…and yes, within months the project resulted in new insights into customer behaviour. But that is another story.

Written by K

June 15, 2016 at 10:00 pm

## A gentle introduction to decision trees using R

with 7 comments

### Introduction

Most techniques of predictive analytics have their origins in probability or statistical theory (see my post on Naïve Bayes, for example).  In this post I’ll look at one that has more a commonplace origin: the way in which humans make decisions.  When making decisions, we typically identify the options available and then evaluate them based on criteria that are important to us.  The intuitive appeal of such a procedure is in no small measure due to the fact that it can be easily explained through a visual. Consider the following graphic, for example:

Figure 1: Example of a simple decision tree (Courtesy: Duncan Hull)

(Original image: https://www.flickr.com/photos/dullhunk/7214525854, Credit: Duncan Hull)

The tree structure depicted here provides a neat, easy-to-follow description of the issue under consideration and its resolution. The decision procedure is based on asking a series of questions, each of which serve to further reduce the domain of possibilities. The predictive technique I discuss in this post,classification and regression trees (CART), works in much the same fashion. It was invented by Leo Breiman and his colleagues in the 1970s.

In what follows, I will use the open source software, R. If you are new to R,   you may want to follow this link for more on the basics of setting up and installing it. Note that the R implementation of the CART algorithm is called RPART (Recursive Partitioning And Regression Trees). This is essentially because Breiman and Co. trademarked the term CART. As some others have pointed out, it is somewhat ironical that the algorithm is now commonly referred to as RPART rather than by the term coined by its inventors.

### A bit about the algorithm

The rpart algorithm works by splitting the dataset recursively, which means that the subsets that arise from a split are further split until a predetermined termination criterion is reached.  At each step, the split is made based on the independent variable that results in the largest possible reduction in heterogeneity of the dependent (predicted) variable.

Splitting rules can be constructed in many different ways, all of which are based on the notion of impurity-  a measure of the degree of heterogeneity of the leaf nodes. Put another way, a leaf node that contains a single class is homogeneous and has impurity=0.   There are three popular impurity quantification methods: Entropy (aka information gain), Gini Index and Classification Error.  Check out this article for a simple explanation of the three methods.

The rpart algorithm offers the entropy  and Gini index methods as choices. There is a fair amount of fact and opinion on the Web about which method is better. Here are some of the better articles I’ve come across:

https://www.quora.com/Are-gini-index-entropy-or-classification-error-measures-causing-any-difference-on-Decision-Tree-classification

http://stats.stackexchange.com/questions/130155/when-to-use-gini-impurity-and-when-to-use-information-gain

https://www.garysieling.com/blog/sklearn-gini-vs-entropy-criteria

http://www.salford-systems.com/resources/whitepapers/114-do-splitting-rules-really-matter

The answer as to which method is the best is: it depends.  Given this, it may be prudent to try out a couple of methods and pick the one that works best for your problem.

Regardless of the method chosen, the splitting rules partition the decision space (a fancy word for the entire dataset) into rectangular regions each of which correspond to a split. Consider the following simple example with two predictors x1 and x2. The first split is at x1=1 (which splits the decision space into two regions x11), the second at x2=2, which splits the (x1>1) region into 2 sub-regions, and finally x1=1.5 which splits the (x1>1,x2>2) sub-region further.

Figure 2: Example of partitioning

It is important to note that the algorithm works by making the best possible choice at each particular stage, without any consideration of whether those choices remain optimal in future stages. That is, the algorithm makes a locally optimal decision at each stage. It is thus quite possible that such a choice at one stage turns out to be sub-optimal in the overall scheme of things.  In other words,  the algorithm does not find a globally optimal tree.

Another important point relates to well-known bias-variance tradeoff in machine learning, which in simple terms is a tradeoff between the degree to which a model fits the training data and its predictive accuracy.  This refers to the general rule that beyond a point, it is counterproductive to improve the fit of a model to the training data as this increases the likelihood of overfitting.  It is easy to see that deep trees are more likely to overfit the data than shallow ones. One obvious way to control such overfitting is to construct shallower trees by stopping the algorithm at an appropriate point based on whether a split significantly improves the fit.  Another is to grow a tree unrestricted and then prune it back using an appropriate criterion. The rpart algorithm takes the latter approach.

Here is how it works in brief:

Essentially one minimises the cost,  $C_{\alpha}(T)$, a quantity that is a  linear combination of the error (essentially, the fraction of misclassified instances, or variance in the case of a continuous variable), $R(T)$  and the number of leaf nodes in the tree, $|\tilde{T} |$:

$C_{\alpha}(T) = R(T) + \alpha |\tilde{T} |$

First, we note that when $\alpha = 0$, this simply returns the original fully grown tree. As $\alpha$ increases, we incur a penalty that is proportional to the number of leaf nodes.  This tends to cause the minimum cost to occur for a tree that is a subtree of the original one (since a subtree will have a smaller number of leaf nodes). In practice we vary $\alpha$ and pick the value that gives the subtree that results in the smallest cross-validated prediction error.  One does not have to worry about programming this because the rpart algorithm actually computes the errors for different values of $\alpha$ for us. All we need to do is pick the value of the coefficient that gives the lowest cross-validated error. I will illustrate this in detail in the next section.

An implication of their tendency to overfit data is that decision trees tend to be sensitive to relatively minor changes in the training datasets. Indeed, small differences can lead to radically different looking trees. Pruning addresses this to an extent, but does not resolve it completely.  A better resolution is offered by the so-called ensemble methods that average over many differently constructed trees. I’ll discuss one such method at length in a future post.

Finally, I should also mention that decision trees can be used for both classification and regression problems (i.e. those in which the predicted variable is discrete and continuous respectively).  I’ll demonstrate both types of problems in the next two sections.

### Classification trees using rpart

To demonstrate classification trees, we’ll use the Ionosphere dataset available in the mlbench package in R. I have chosen this dataset because it nicely illustrates the points I wish to make in this post. In general, you will almost always find that algorithms that work fine on classroom datasets do not work so well in the real world…but of course, you know that already!

We begin by setting the working directory, loading the required packages (rpart and mlbench) and then loading the Ionosphere dataset.

#set working directory if needed (modify path as needed)
setwd(“C:/Users/Kailash/Documents/decisiontrees”)
#load required libraries – rpart for classification and regression trees
library(rpart)
#mlbench for Ionosphere dataset
library(mlbench)
#load Ionosphere
data(“Ionosphere”)

Next we separate the data into training and test sets. We’ll use the former to build the model and the latter to test it. To do this, I use a simple scheme wherein I randomly select 80% of the data for the training set and assign the remainder to the test data set. This is easily done in a single R statement that invokes the uniform distribution (runif) and the vectorised function, ifelse. Before invoking runif, I set a seed integer to my favourite integer in order to ensure reproducibility of results.

#set seed to ensure reproducible results
set.seed(42)
#split into training and test sets
Ionosphere[,”train”] <- ifelse(runif(nrow(Ionosphere))<0.8,1,0)
#separate training and test sets
trainset <- Ionosphere[Ionosphere$train==1,] testset <- Ionosphere[Ionosphere$train==0,]
#get column index of train flag
trainColNum <- grep("train",names(trainset))
#remove train flag column from train and test sets
trainset <- trainset[,-trainColNum]
testset <- testset[,-trainColNum]

In the above, I have also removed the training flag from the training and test datasets.

Next we  invoke rpart. I strongly recommend you take some time to go through the documentation and understand the parameters and their defaults values.  Note that we need to remove the predicted variable from the dataset before passing the latter on to the algorithm, which is why we need to find the column index of the  predicted variable (first line below). Also note that we set the method parameter to “class“, which simply tells the algorithm that the predicted variable is discrete.  Finally, rpart uses Gini rule for splitting by default, and we’ll stick with this option.

#get column index of predicted variable in dataset
typeColNum <- grep("Class",names(Ionosphere))
#build model
rpart_model <- rpart(Class~.,data = trainset, method="class")
#plot tree
plot(rpart_model);text(rpart_model)

The resulting plot is shown in Figure 3 below.  It is  quite self-explanatory so I  won’t dwell on it here.

Figure 3: A classification tree for Ionosphere dataset

Next we see how good the model is by seeing how it fares against the test data.

#…and the moment of reckoning
rpart_predict <- predict(rpart_model,testset[,-typeColNum],type="class")
mean(rpart_predict==testset$Class) [1] 0.8450704 #confusion matrix table(pred=rpart_predict,true=testset$Class)
 pred true bad good bad 17 2 good 9 43

Note that we need to verify the above results by doing multiple runs, each using different training and test sets. I will  do this later, after discussing pruning.

Next, we prune the tree using the cost complexity criterion. Basically, the intent is to see if a shallower subtree can give us comparable results. If so, we’d be better of choosing the shallower tree because it reduces the likelihood of overfitting.

As described earlier, we choose the appropriate pruning parameter (aka cost-complexity parameter) $\alpha$ by picking the value that results in the lowest prediction error. Note that all relevant computations have already been carried out by R when we built the original tree (the call to rpart in the code above). All that remains now is to pick the value of $\alpha$:

#cost-complexity pruning
printcp(rpart_model)
 CP nsplit rel error xerror xstd 1 0.57 0 1.00 1.00 0.080178 2 0.20 1 0.43 0.46 0.062002 3 0.02 2 0.23 0.26 0.048565 4 0.01 4 0.19 0.35

It is clear from the above, that the lowest cross-validation error (xerror in the table) occurs for $\alpha =0.02$ (this is CP in the table above).   One can find CP programatically like so:

# get index of CP with lowest xerror
opt <- which.min(rpart_model$cptable[,"xerror"]) #get its value cp <- rpart_model$cptable[opt, "CP"]

Next, we prune the tree based on this value of CP:

#prune tree
pruned_model <- prune(rpart_model,cp)
#plot tree
plot(pruned_model);text(pruned_model)

Note that rpart will use a default CP value of 0.01 if you don’t specify one in prune.

The pruned tree is shown in Figure 4 below.

Figure 4: A pruned classification tree for Ionosphere dataset

Let’s see how this tree stacks up against the fully grown one shown in Fig 3.

#find proportion of correct predictions using test set
rpart_pruned_predict <- predict(pruned_model,testset[,-typeColNum],type="class")
mean(rpart_pruned_predict==testset$Class) [1] 0.8873239 This seems like an improvement over the unpruned tree, but one swallow does not a summer make. We need to check that this holds up for different training and test sets. This is easily done by creating multiple random partitions of the dataset and checking the efficacy of pruning for each. To do this efficiently, I’ll create a function that takes the training fraction, number of runs (partitions) and the name of the dataset as inputs and outputs the proportion of correct predictions for each run. It also optionally prunes the tree. Here’s the code: #function to do multiple runs multiple_runs_classification <- function(train_fraction,n,dataset,prune_tree=FALSE){ fraction_correct <- rep(NA,n) set.seed(42) for (i in 1:n){ dataset[,”train”] <- ifelse(runif(nrow(dataset))<0.8,1,0) trainColNum <- grep("train",names(dataset)) typeColNum <- grep("Class",names(dataset)) trainset <- dataset[dataset$train==1,-trainColNum]
testset <- dataset[dataset$train==0,-trainColNum] rpart_model <- rpart(Class~.,data = trainset, method="class") if(prune_tree==FALSE) { rpart_test_predict <- predict(rpart_model,testset[,-typeColNum],type="class") fraction_correct[i] <- mean(rpart_test_predict==testset$Class)
}else{
opt <- which.min(rpart_model$cptable[,"xerror"]) cp <- rpart_model$cptable[opt, "CP"]
pruned_model <- prune(rpart_model,cp)
rpart_pruned_predict <- predict(pruned_model,testset[,-typeColNum],type="class")
fraction_correct[i] <- mean(rpart_pruned_predict==testset$Class) } } return(fraction_correct) } Note that in the above, I have set the default value of the prune_tree to FALSE, so the function will execute the first branch of the if statement unless the default is overridden. OK, so let’s do 50 runs with and without pruning, and check the mean and variance of the results for both sets of runs. #50 runs, no pruning unpruned_set <- multiple_runs_classification(0.8,50,Ionosphere) mean(unpruned_set) [1] 0.8772763 sd(unpruned_set) [1] 0.03168975 #50 runs, with pruning pruned_set <- multiple_runs_classification(0.8,50,Ionosphere,prune_tree=TRUE) mean(pruned_set) [1] 0.9042914 sd(pruned_set) [1] 0.02970861 So we see that there is an improvement of about 3% with pruning. Also, if you were to plot the trees as we did earlier, you would see that this improvement is achieved with shallower trees. Again, I point out that this is not always the case. In fact, it often happens that pruning results in worse predictions, albeit with better reliability – a classic illustration of the bias-variance tradeoff. ### Regression trees using rpart In the previous section we saw how one can build decision trees for situations in which the predicted variable is discrete. Let’s now look at the case in which the predicted variable is continuous. We’ll use the Boston Housing dataset from the mlbench package. Much of the discussion of the earlier section applies here, so I’ll just display the code, explaining only the differences. #load Boston Housing dataset data(“BostonHousing”) #set seed to ensure reproducible results set.seed(42) #split into training and test sets BostonHousing[,”train”] <- ifelse(runif(nrow(BostonHousing))<0.8,1,0) #separate training and test sets trainset <- BostonHousing[BostonHousing$train==1,]
testset <- BostonHousing[BostonHousing$train==0,] #get column index of train flag trainColNum <- grep("train",names(trainset)) #remove train flag column from train and test sets trainset <- trainset[,-trainColNum] testset <- testset[,-trainColNum] Next we invoke rpart, noting that the predicted variable is medv (median value of owner-occupied homes in$1000 units) and that we need to set the method parameter to “anova“. The latter tells rpart that the predicted variable is continuous (i.e that this is a regression problem).

#build model
rpart_model <- rpart(medv~.,data = trainset, method="anova")
#plot tree
plot(rpart_model);text(rpart_model)

The plot of the tree is shown in Figure 5 below.

Figure 5: A regression tree for Boston Housing dataset

Next, we need to see how good the predictions are. Since the dependent variable is continuous, we cannot compare the predictions directly against the test set. Instead, we calculate the root mean square (RMS) error. To do this, we request rpart to output the predictions as a vector – one prediction per record in the test dataset. The RMS error can then easily be calculated by comparing this vector with the medv column in the test dataset.

Here is the relevant code:

#…the moment of reckoning
rpart_test_predict <- predict(rpart_model,testset[,-resultColNum],type = "vector" )
#calculate RMS error
rmsqe <- sqrt(mean((rpart_test_predict-testset$medv)^2))) rmsqe [1] 4.586388 Again, we need to do multiple runs to check on the reliability of the predictions. However, you already know how to do that so I will leave it to you. Moving on, we prune the tree using the cost complexity criterion as before. The code is exactly the same as in the classification problem. # get index of CP with lowest xerror opt <- which.min(rpart_model$cptable[,"xerror"])
#get its value
cp <- rpart_model$cptable[opt, "CP"] #prune tree pruned_model <- prune(rpart_model,cp) #plot tree plot(pruned_model);text(pruned_model) The tree is unchanged so I won’t show it here. This means, as far as the cost complexity pruning is concerned, the optimal subtree is the same as the original tree. To confirm this, we’d need to do multiple runs as before – something that I’ve already left as as an exercise for you :). Basically, you’ll need to write a function analogous to the one above, that computes the root mean square error instead of the proportion of correct predictions. ### Wrapping up This brings us to the end of my introduction to classification and regression trees using R. Unlike some articles on the topic I have attempted to describe each of the steps in detail and provide at least some kind of a rationale for them. I hope you’ve found the description and code snippets useful. I’ll end by reiterating a couple points I made early in this piece. The nice thing about decision trees is that they are easy to explain to the users of our predictions. This is primarily because they reflect the way we think about how decisions are made in real life – via a set of binary choices based on appropriate criteria. That said, in many practical situations decision trees turn out to be unstable: small changes in the dataset can lead to wildly different trees. It turns out that this limitation can be addressed by building a variety of trees using different starting points and then averaging over them. This is the domain of the so-called random forest algorithm.We’ll make the journey from decision trees to random forests in a future post. Postscript, 20th September 2016: I finally got around to finishing my article on random forests. Written by K February 16, 2016 at 6:33 pm ## A gentle introduction to network graphs using R and Gephi with 7 comments ### Introduction Graph theory is the an area of mathematics that analyses relationships between pairs of objects. Typically graphs consist of nodes (points representing objects) and edges (lines depicting relationships between objects). As one might imagine, graphs are extremely useful in visualizing relationships between objects. In this post, I provide a detailed introduction to network graphs using R, the premier open source tool statistics package for calculations and the excellent Gephi software for visualization. The article is organised as follows: I begin by defining the problem and then spend some time developing the concepts used in constructing the graph Following this, I do the data preparation in R and then finally build the network graph using Gephi. ### The problem In an introductory article on cluster analysis, I provided an in-depth introduction to a couple of algorithms that can be used to categorise documents automatically. Although these techniques are useful, they do not provide a feel for the relationships between different documents in the collection of interest. In the present piece I show network graphs can be used to to visualise similarity-based relationships within a corpus. ### Document similarity There are many ways to quantify similarity between documents. A popular method is to use the notion of distance between documents. The basic idea is simple: documents that have many words in common are “closer” to each other than those that share fewer words. The problem with distance, however, is that it can be skewed by word count: documents that have an unusually high word count will show up as outliers even though they may be similar (in terms of words used) to other documents in the corpus. For this reason, we will use another related measure of similarity that does not suffer from this problem – more about this in a minute. Representing documents mathematically As I explained in my article on cluster analysis, a document can be represented as a point in a conceptual space that has dimensionality equal to the number of distinct words in the collection of documents. I revisit and build on that explanation below. Say one has a simple document consisting of the words “five plus six”, one can represent it mathematically in a 3 dimensional space in which the individual words are represented by the three axis (See Figure 1). Here each word is a coordinate axis (or dimension). Now, if one connects the point representing the document (point A in the figure) to the origin of the word-space, one has a vector, which in this case is a directed line connecting the point in question to the origin. Specifically, the point A can be represented by the coordinates $(1, 1, 1)$ in this space. This is a nice quantitative representation of the fact that the words five, plus and one appear in the document exactly once. Note, however, that we’ve assumed the order of words does not matter. This is a reasonable assumption in some cases, but not always so. Figure 1 As another example consider document, B, which consists of only two words: “five plus” (see Fig 2). Clearly this document shares some similarity with document but it is not identical. Indeed, this becomes evident when we note that document (or point) B is simply the point$latex(1, 1, 0)$in this space, which tells us that it has two coordinates (words/frequencies) in common with document (or point) A. Figure 2 To be sure, in a realistic collection of documents we would have a large number of distinct words, so we’d have to work in a very high dimensional space. Nevertheless, the same principle holds: every document in the corpus can be represented as a vector consisting of a directed line from the origin to the point to which the document corresponds. Cosine similarity Now it is easy to see that two documents are identical if they correspond to the same point. In other words, if their vectors coincide. On the other hand, if they are completely dissimilar (no words in common), their vectors will be at right angles to each other. What we need, therefore, is a quantity that varies from 0 to 1 depending on whether two documents (vectors) are dissimilar(at right angles to each other) or similar (coincide, or are parallel to each other). Now here’s the ultra-cool thing, from your high school maths class, you know there is a trigonometric ratio which has exactly this property – the cosine! What’s even cooler is that the cosine of the angle between two vectors is simply the dot product of the two vectors, which is sum of the products of the individual elements of the vector, divided by the product of the lengths of the two vectors. In three dimensions this can be expressed mathematically as: $\cos(\theta)= \displaystyle \frac{x_1 x_2+y_1 y_2+z_1 z_2}{\sqrt{x_1^2+y_1^2+z_1^2}\sqrt{x_2^2+y_2^2+z_2^2}}...(1)$ where the two vectors are $(x_{1},y_{1},z_{1})$ and $(x_{2},y_{2},z_{2})$, and $\theta$ is the angle between the two vectors (see Fig 2). The upshot of the above is that the cosine of the angle between the vector representation of two documents is a reasonable measure of similarity between them. This quantity, sometimes referred to as cosine similarity, is what we’ll take as our similarity measure in the rest of this article. ### The adjacency matrix If we have a collection of $N$ documents, we can calculate the similarity between every pair of documents as we did for A and B in the previous section. This would give us a set of $N^2$ numbers between 0 and 1, which can be conveniently represented as a matrix. This is sometimes called the adjacency matrix. Beware, though, this term has many different meanings in the math literature. I use it in the sense specified above. Since every document is identical to itself, the diagonal elements of the matrix will all be 1. These similarities are trivial (we know that every document is identical to itself!) so we’ll set the diagonal elements to zero. Another important practical point is that visualizing every relationship is going to make a very messy graph. There would be $N(N-1)$ edges in such a graph, which would make it impossible to make sense of if we have more than a handful of documents. For this reason, it is normal practice to choose a cutoff value of similarity below which it is set to zero. ### Building the adjacency matrix using R We now have enough background to get down to the main point of this article – visualizing relationships between documents. The first step is to build the adjacency matrix. In order to do this, we have to build the document term matrix (DTM) for the collection of documents, a process which I have dealt with at length in my introductory pieces on text mining and topic modeling. In fact, the steps are actually identical to those detailed in the second piece. I will therefore avoid lengthy explanations here. However, I’ve listed all the code below with brief comments (for those who are interested in trying this out, the document corpus can be downloaded here and a pdf listing of the R code can be obtained here.) OK, so here’s the code listing: #load text mining library library(tm) #set working directory (modify path as needed) setwd(“C:\\Users\\Kailash\\Documents\\TextMining”) #load files into corpus #get listing of .txt files in directory filenames <- list.files(getwd(),pattern=”*.txt”) #read files into a character vector files <- lapply(filenames,readLines) #create corpus from vector docs <- Corpus(VectorSource(files)) #inspect a particular document in corpus writeLines(as.character(docs[[30]])) #start preprocessing #Transform to lower case docs <-tm_map(docs,content_transformer(tolower)) #remove potentially problematic symbols toSpace <- content_transformer(function(x, pattern) { return (gsub(pattern, ” “, x))}) docs <- tm_map(docs, toSpace, “-“) docs <- tm_map(docs, toSpace, “’”) docs <- tm_map(docs, toSpace, “‘”) docs <- tm_map(docs, toSpace, “•”) docs <- tm_map(docs, toSpace, “””) docs <- tm_map(docs, toSpace, ““”) #remove punctuation docs <- tm_map(docs, removePunctuation) #Strip digits docs <- tm_map(docs, removeNumbers) #remove stopwords docs <- tm_map(docs, removeWords, stopwords(“english”)) #remove whitespace docs <- tm_map(docs, stripWhitespace) #Good practice to check every now and then writeLines(as.character(docs[[30]])) #Stem document docs <- tm_map(docs,stemDocument) #fix up 1) differences between us and aussie english 2) general errors docs <- tm_map(docs, content_transformer(gsub), pattern = “organiz”, replacement = “organ”) docs <- tm_map(docs, content_transformer(gsub), pattern = “organis”, replacement = “organ”) docs <- tm_map(docs, content_transformer(gsub), pattern = “andgovern”, replacement = “govern”) docs <- tm_map(docs, content_transformer(gsub), pattern = “inenterpris”, replacement = “enterpris”) docs <- tm_map(docs, content_transformer(gsub), pattern = “team-“, replacement = “team”) #define and eliminate all custom stopwords myStopwords <- c(“can”, “say”,”one”,”way”,”use”, “also”,”howev”,”tell”,”will”, “much”,”need”,”take”,”tend”,”even”, “like”,”particular”,”rather”,”said”, “get”,”well”,”make”,”ask”,”come”,”end”, “first”,”two”,”help”,”often”,”may”, “might”,”see”,”someth”,”thing”,”point”, “post”,”look”,”right”,”now”,”think”,”‘ve “, “‘re “,”anoth”,”put”,”set”,”new”,”good”, “want”,”sure”,”kind”,”larg”,”yes,”,”day”,”etc”, “quit”,”sinc”,”attempt”,”lack”,”seen”,”awar”, “littl”,”ever”,”moreov”,”though”,”found”,”abl”, “enough”,”far”,”earli”,”away”,”achiev”,”draw”, “last”,”never”,”brief”,”bit”,”entir”,”brief”, “great”,”lot”) docs <- tm_map(docs, removeWords, myStopwords) #inspect a document as a check writeLines(as.character(docs[[30]])) #Create document-term matrix dtm <- DocumentTermMatrix(docs) The rows of a DTM are document vectors akin to the vector representations of documents A and B discussed earlier. The DTM therefore contains all the information we need to calculate the cosine similarity between every pair of documents in the corpus (via equation 1). The R code below implements this, after taking care of a few preliminaries. #convert dtm to matrix m<-as.matrix(dtm) #write as csv file write.csv(m,file=”dtmEight2Late.csv”) #Map filenames to matrix row numbers #these numbers will be used to reference #files in the network graph filekey <- cbind(rownames(m),filenames) write.csv(filekey,”filekey.csv”) #compute cosine similarity between document vectors #converting to distance matrix sets diagonal elements to 0 cosineSim <- function(x){ as.dist(x%*%t(x)/(sqrt(rowSums(x^2) %*% t(rowSums(x^2))))) } cs <- cosineSim(m) write.csv(as.matrix(cs),file=”csEight2Late.csv”) #adjacency matrix: set entries below a certain threshold to 0. #We choose half the magnitude of the largest element of the matrix #as the cutoff. This is an arbitrary choice cs[cs < max(cs)/2] <- 0 cs <- round(cs,3) write.csv(as.matrix(cs),file=”AdjacencyMatrix.csv”) A few lines need a brief explanation: First up, although the DTM is a matrix, it is internally stored in a special form suitable for sparse matrices. We therefore have to explicitly convert it into a proper matrix before using it to calculate similarity. Second, the names I have given the documents are way too long to use as labels in the network diagram. I have therefore mapped the document names to the row numbers which we’ll use in our network graph later. The mapping back to the original document names is stored in filekey.csv. For future reference, the mapping is shown in Table 1 below.  File number Name 1 BeyondEntitiesAndRelationships.txt 2 bigdata.txt 3 ConditionsOverCauses.txt 4 EmergentDesignInEnterpriseIT.txt 5 FromInformationToKnowledge.txt 6 FromTheCoalface.txt 7 HeraclitusAndParmenides.txt 8 IroniesOfEnterpriseIT.txt 9 MakingSenseOfOrganizationalChange.txt 10 MakingSenseOfSensemaking.txt 11 ObjectivityAndTheEthicalDimensionOfDecisionMaking.txt 12 OnTheInherentAmbiguitiesOfManagingProjects.txt 13 OrganisationalSurprise.txt 14 ProfessionalsOrPoliticians.txt 15 RitualsInInformationSystemDesign.txt 16 RoutinesAndReality.txt 17 ScapegoatsAndSystems.txt 18 SherlockHolmesFailedProjects.txt 19 sherlockHolmesMgmtFetis.txt 20 SixHeresiesForBI.txt 21 SixHeresiesForEnterpriseArchitecture.txt 22 TheArchitectAndTheApparition.txt 23 TheCloudAndTheGrass.txt 24 TheConsultantsDilemma.txt 25 TheDangerWithin.txt 26 TheDilemmasOfEnterpriseIT.txt 27 TheEssenceOfEntrepreneurship.txt 28 ThreeTypesOfUncertainty.txt 29 TOGAFOrNotTOGAF.txt 30 UnderstandingFlexibility.txt Table 1: File mappings Finally, the distance function (as.dist) in the cosine similarity function sets the diagonal elements to zero because the distance between a document and itself is zero…which is just a complicated way of saying that a document is identical to itself 🙂 The last three lines of code above simply implement the cutoff that I mentioned in the previous section. The comments explain the details so I need say no more about it. …which finally brings us to Gephi. ### Visualizing document similarity using Gephi Gephi is an open source, Java based network analysis and visualisation tool. Before going any further, you may want to download and install it. While you’re at it you may also want to download this excellent quick start tutorial. Go on, I’ll wait for you… To begin with, there’s a little formatting quirk that we need to deal with. Gephi expects separators in csv files to be semicolons (;) . So, your first step is to open up the adjacency matrix that you created in the previous section (AdjacencyMatrix.csv) in a text editor and replace commas with semicolons. Once you’ve done that, fire up Gephi, go to File > Open, navigate to where your Adjacency matrix is stored and load the file. If it loads successfully, you should see a feedback panel as shown in Figure 3. By default Gephi creates a directed graph (i.e one in which the edges have arrows pointing from one node to another). Change this to undirected and click OK. Figure 3: Gephi import feedback Once that is done, click on overview (top left of the screen). You should end up with something like Figure 4. Figure 4: Initial overview after loading adjacency matrix Gephi has sketched out an initial network diagram which depicts the relationships between documents…but it needs a bit of work to make it look nicer and more informative. The quickstart tutorial mentioned earlier describes various features that can be used to manipulate and prettify the graph. In the remainder of this section, I list some that I found useful. Gephi offers many more. Do explore, there’s much more than I can cover in an introductory post. First some basics. You can: • Zoom and pan using mouse wheel and right button. • Adjust edge thicknesses using the slider next to text formatting options on bottom left of main panel. • Re-center graph via the magnifying glass icon on left of display panel (just above size adjuster). • Toggle node labels on/off by clicking on grey T symbol on bottom left panel. Figure 5 shows the state of the diagram after labels have been added and edge thickness adjusted (note that your graph may vary in appearance). Figure 5: graph with node labels and adjusted edge thicknesses The default layout of the graph is ugly and hard to interpret. Let’s work on fixing it up. To do this, go over to the layout panel on the left. Experiment with different layouts to see what they do. After some messing around, I found the Fruchtermann-Reingold and Force Atlas options to be good for this graph. In the end I used Force Atlas with a Repulsion Strength of 2000 (up from the default of 200) and an Attraction Strength of 1 (down from the default of 10). I also adjusted the figure size and node label font size from the graph panel in the center. The result is shown in Figure 6. Figure 6: Graph after using Force Atlas layout This is much better. For example, it is now evident that document 9 is the most connected one (which table 9 tells us is a transcript of a conversation with Neil Preston on organisational change). It would be nice if we could colour code edges/nodes and size nodes by their degree of connectivity. This can be done via the ranking panel above the layout area where you’ve just been working. In the Nodes tab select Degree as the rank parameter (this is the degree of connectivity of the node) and hit apply. Select your preferred colours via the small icon just above the colour slider. Use the colour slider to adjust the degree of connectivity at which colour transitions occur. Do the same for edges, selecting weight as the rank parameter(this is the degree of similarity between the two douments connected by the edge). With a bit of playing around, I got the graph shown in the screenshot below (Figure 7). Figure 5: Connectivity-based colouring of edges and nodes. If you want to see numerical values for the rankings, hit the results list icon on the bottom left of the ranking panel. You can see numerical ranking values for both nodes and edges as shown in Figures 8 and 9. Figure 8: Node ranking (see left of figure) Figure 9: Edge ranking It is easy to see from the figure that documents 21 and 29 are the most similar in terms of cosine ranking. This makes sense, they are pieces in which I have ranted about the current state of enterprise architecture – the first article is about EA in general and the other about the TOGAF framework. If you have a quick skim through, you’ll see that they have a fair bit in common. Finally, it would be nice if we could adjust node size to reflect the connectedness of the associated document. You can do this via the “gem” symbol on the top right of the ranking panel. Select appropriate min and max sizes (I chose defaults) and hit apply. The node size is now reflective of the connectivity of the node – i.e. the number of other documents to which it is cosine similar to varying degrees. The thickness of the edges reflect the degree of similarity. See Figure 10. Figure 10: Node sizes reflecting connectedness Now that looks good enough to export. To do this, hit the preview tab on main panel and make following adjustments to the default settings: Under Node Labels: 1. Check Show Labels 2. Uncheck proportional size 3. Adjust font to required size Under Edges: 1. Change thickness to 10 2. Check rescale weight Hit refresh after making the above adjustments. You should get something like Fig 11. Figure 11: Export preview All that remains now is to do the deed: hit export SVG/PDF/PNG to export the diagram. My output is displayed in Figure 12. It clearly shows the relationships between the different documents (nodes) in the corpus. The nodes with the highest connectivity are indicated via node size and colour (purple for high, green for low) and strength of similarity is indicated by edge thickness. Figure 12: Gephi network graph of document corpus …which brings us to the end of this journey. ### Wrapping up The techniques of text analysis enable us to quantify relationships between documents. Document similarity is one such relationship. Numerical measures are good, but the comprehensibility of these can be further enhanced through meaningful visualisations. Indeed, although my stated objective in this article was to provide an introduction to creating network graphs using Gephi and R (which I hope I’ve succeeded in doing), a secondary aim was to show how document similarity can be quantified and visualised. I sincerely hope you’ve found the discussion interesting and useful. Many thanks for reading! As always, your feedback would be greatly appreciated. Written by K December 2, 2015 at 7:20 am ## A gentle introduction to Naïve Bayes classification using R with 14 comments ### Preamble One of the key problems of predictive analytics is to classify entities or events based on a knowledge of their attributes. An example: one might want to classify customers into two categories, say, ‘High Value’ or ‘Low Value,’ based on a knowledge of their buying patterns. Another example: to figure out the party allegiances of representatives based on their voting records. And yet another: to predict the species a particular plant or animal specimen based on a list of its characteristics. Incidentally, if you haven’t been there already, it is worth having a look at Kaggle to get an idea of some of the real world classification problems that people tackle using techniques of predictive analytics. Given the importance of classification-related problems, it is no surprise that analytics tools offer a range of options. My favourite (free!) tool, R, is no exception: it has a plethora of state of the art packages designed to handle a wide range of problems. One of the problems with this diversity of choice is that it is often confusing for beginners to figure out which one to use in a particular situation. Over the next several months, I intend to write up tutorial articles covering many of the common algorithms, with a particular focus on their strengths and weaknesses; explaining where they work well and where they don’t. I’ll kick-off this undertaking with a simple yet surprisingly effective algorithm – the Naïve Bayes classifier. ### Just enough theory I’m going to assume you have R and RStudio installed on your computer. If you need help with this, please follow the instructions here. To introduce the Naive Bayes algorithm, I will use the HouseVotes84 dataset, which contains US congressional voting records for 1984. The data set is in the mlbench package which is not part of the base R installation. You will therefore need to install it if you don’t have it already. Package installation is a breeze in RStudio – just go to Tools > Install Packages and follow the prompts. The HouseVotes84 dataset describes how 435 representatives voted – yes (y), no (n) or unknown (NA) – on 16 key issues presented to Congress. The dataset also provides the party affiliation of each representative – democrat or republican. Let’s begin by exploring the dataset. To do this, we load mlbench, fetch the dataset and get some summary stats on it. (Note: a complete listing of the code in this article can be found here) #load mlbench library library(mlbench) #set working directory if needed (modify path as needed) setwd(“C:/Users/Kailash/Documents/NaiveBayes”) #load HouseVotes84 dataset data(“HouseVotes84”) It is good to begin by exploring the data visually. To this end, let’s do some bar plots using the basic graphic capabilities of R: #barplots for specific issue plot(as.factor(HouseVotes84[,2])) title(main=”Votes cast for issue”, xlab=”vote”, ylab=”# reps”) #by party plot(as.factor(HouseVotes84[HouseVotes84$Class==’republican’,2]))
title(main=”Republican votes cast for issue 1″, xlab=”vote”, ylab=”# reps”)
plot(as.factor(HouseVotes84[HouseVotes84$Class==’democrat’,2])) title(main=”Democrat votes cast for issue 1″, xlab=”vote”, ylab=”# reps”) The plots are shown in Figures 1 through 3. Fig 1: y and n votes for issue 1 Fig 2: Republican votes for issue 1. Fig 3: Democrat votes for issue 1. Among other things, such plots give us a feel for the probabilities associated with how representatives from parties tend to vote on specific issues. The classification problem at hand is to figure out the party affiliation from a knowledge of voting patterns. For simplicity let us assume that there are only 3 issues voted on instead of the 16 in the actual dataset. In concrete terms we wish to answer the question, “what is the probability that a representative is, say, a democrat (D) given that he or she has voted, say, $(v1 = y, v2=n,v3 = y)$ on the three issues?” To keep things simple I’m assuming there are no NA values. In the notation of conditional probability this can be written as, $P(D|v1=y, v2=n,v3=y)$ (Note: If you need a refresher on conditional probability, check out this post for a simple explanation.) By Bayes theorem, which I’ve explained at length in this post, this can be recast as, $P(D|v1=y, v2=n,v3=y) = \displaystyle \frac{p(D) p(v1=y, v2=n,v3=y|D)}{p(v1=y, v2=n,v3=y)}......(1)$ We’re interested only in relative probabilities of the representative being a democrat or republican because the predicted party affiliation depends only on which of the two probabilities is larger (the actual value of the probability is not important). This being the case, we can factor out any terms that are constant. As it happens, the denominator of the above equation – the probability of a particular voting pattern – is a constant because it depends on the total number of representatives (from both parties) who voted a particular way. Now, using the chain rule of conditional probability, we can rewrite the numerator as: $p(D) p(v1=y, v2=n,v3=y|D)$ $= p(D)p(v1=y|D) p(v2=n,v3=y|D,v1=y)$ Basically, the second term on the left hand side, $p(v1=y, v2=n,v3=y|D)$, is the probability of getting a particular voting pattern (y,n,y) assuming the rep is a Democrat (D). The definition of conditional probability allows us to rewrite this as the probability of getting a n vote for issue v2 and a y vote for issue v3 given that the rep is a Democrat who has voted y on issue v1. Again, this is simply a consequence of the definition of conditional probability. Another application of the chain rule gives: $p(D) p(v1=y, v2=n,v3=y|D)$ $= p(D)p(v1=y|d) p(v2=n|D,v1=y) p(v3=y|D,v1=y,v2=n)$ Where we have now factored out the n vote on the second issue. The key assumption of Naïve Bayes is that the conditional probability of each feature given the class is independent of all other features. In mathematical terms this means that, $p(v2=n|D,v1=y) = p(v2=n|D)$ and $p(v3=y|D,v1=y,v2=n) = p(v3=y|D)$ The quantity of interest, the numerator of equation (1) can then be written as: $p(D) p(v1=y, v2=n,v3=y|D)$ $= p(D)p(v1=y|D)p(v2=n|D)p(v3=y|D).......(2)$ The assumption of independent conditional probabilities is a drastic one. What it is saying is that the features are completely independent of each other. This is clearly not the case in the situation above: how representatives vote on a particular issue is coloured by their beliefs and values. For example, the conditional probability of voting patterns on socially progressive issues are definitely not independent of each other. However, as we shall see in the next section, the Naïve Bayes assumption works well for this problem as it does in many other situations where we know upfront that it is grossly incorrect. Another good example of the unreasonable efficacy of Naive Bayes is in spam filtering. In the case of spam, the features are individual words in an email. It is clear that certain word combinations tend to show up consistently in spam – for example, “online”, “meds”, “Viagra” and “pharmacy.” In other words, we know upfront that their occurrences are definitely not independent of each other. Nevertheless, Naïve Bayes based spam detectors which assume mutual independence of features do remarkably well in distinguishing spam from ham. Why is this so? To explain why, I return to a point I mentioned earlier: to figure out the affiliation associated with a particular voting pattern (say, v1=y, v2=n,v3=y) one only needs to know which of the two probabilities $p(R| v1=y, v2=n,v3=y)$ and $p(D| V1=y, V2=n, V3=y)$ is greater than the other. That is, the values of these probabilities are not important in determining the party affiliations. This hints as to why the independence assumption might not be so quite so idiotic. Since the prediction depends only the on the maximum, the algorithm will get it right even if there are dependencies between feature providing the dependencies do not change which class has the maximum probability (once again, note that only the maximal class is important here, not the value of the maximum). Yet another reason for the surprising success of Naïve Bayes is that dependencies often cancel out across a large set of features. But, of course, there is no guarantee that this will always happen. In general, Naïve Bayes algorithms work better for problems in which the dependent (predicted) variable is discrete, even when there are dependencies between features (spam detection is a good example). They work less well for regression problems – i.e those in which predicted variables are continuous. I hope the above has given you an intuitive feel for how Naïve Bayes algorithms work. I don’t know about you, but my head’s definitely spinning after writing out all that mathematical notation. It’s time to clear our heads by doing some computation. ### Naïve Bayes in action There are a couple of well-known implementations of Naïve Bayes in R. One of them is the naiveBayes method in the e1071 package and the other is NaiveBayes method in the klaR package. I’ll use the former for no other reason than it seems to be more popular. That said, I have used the latter too and can confirm that it works just as well. We’ve already loaded and explored the HouseVotes84 dataset. One of the things you may have noticed when summarising the data is that there are a fair number of NA values. Naïve Bayes algorithms typically handle NA values either by ignoring records that contain any NA values or by ignoring just the NA values. These choices are indicated by the value of the variable na.action in the naiveBayes algorithm, which is set to na.omit (to ignore the record) or na.pass (to ignore the value). Just for fun, we’ll take a different approach. We’ll impute NA values for a given issue and party by looking at how other representatives from the same party voted on the issue. This is very much in keeping with the Bayesian spirit: we infer unknowns based on a justifiable belief – that is, belief based on the evidence. To do this I write two functions: one to compute the number of NA values for a given issue (vote) and class (party affiliation), and the other to calculate the fraction of yes votes for a given issue (column) and class (party affiliation). #Functions needed for imputation #function to return number of NAs by vote and class (democrat or republican) na_by_col_class <- function (col,cls){return(sum(is.na(HouseVotes84[,col]) & HouseVotes84$Class==cls))}
#function to compute the conditional probability that a member of a party will cast a ‘yes’ vote for
#a particular issue. The probability is based on all members of the party who #actually cast a vote on the issue (ignores NAs).
p_y_col_class <- function(col,cls){
sum_y<-sum(HouseVotes84[,col]==’y’ & HouseVotes84$Class==cls,na.rm = TRUE) sum_n<-sum(HouseVotes84[,col]==’n’ & HouseVotes84$Class==cls,na.rm = TRUE)
return(sum_y/(sum_y+sum_n))}
#Check that functions work!
> p_y_col_class(2,’democrat’)
[1] 0.6046512
> p_y_col_class(2,’republican’)
[1] 0.1878788
> na_by_col_class(2,’democrat’)
[1] 9
> na_by_col_class(2,’republican’)
> [1] 3

Before proceeding,  you might want to go back to the data and convince yourself that these values are sensible.

We can now impute the NA values based on the above. We do this by randomly assigning values ( y or n) to NAs, based on the proportion of members of a party who have voted y or n. In practice, we do this by invoking the uniform distribution and setting an NA value to y if the random number returned is less than the probability of a yes vote and to n otherwise. This is not as complicated as it sounds; you should be able to figure the logic out from the code below.

#impute missing values.
for (i in 2:ncol(HouseVotes84)) {
if(sum(is.na(HouseVotes84[,i])>0)) {
c1 <- which(is.na(HouseVotes84[,i])& HouseVotes84$Class==’democrat’,arr.ind = TRUE) c2 <- which(is.na(HouseVotes84[,i])& HouseVotes84$Class==’republican’,arr.ind = TRUE)
HouseVotes84[c1,i] <-
ifelse(runif(na_by_col_class(i,’democrat’))<p_y_col_class(i,’democrat’),’y’,’n’)
HouseVotes84[c2,i] <-
ifelse(runif(na_by_col_class(i,’republican’))<p_y_col_class(i,’republican’),’y’,’n’)}
}

Note that the which function filters  indices by the criteria specified in the arguments and ifelse is a vectorised conditional function which enables us to apply logical criteria to multiple elements of a vector.

At this point it is a good idea to check that the NAs in each column have been set according to the voting patterns of non-NAs for a given party. You can use the p_y_col_class() function to check that the new probabilities are close to the old ones. You might want to do this before you proceed any further.

The next step is to divide the available data into training and test datasets. The former will be used to train the algorithm and produce a predictive model. The effectiveness of the model will then be tested using the test dataset. There is a great deal of science and art behind the creation of training and testing datasets. An important consideration is that both sets must contain records that are representative of the entire dataset. This can be difficult to do, especially when data is scarce and there are predictors that do not vary too much…or vary wildly for that matter. On the other hand, problems can also arise when there are redundant predictors. Indeed, the much of the art of successful prediction lies in figuring out which predictors are likely to lead to better predictions, an area known as feature selection. However, that’s a topic for another time. Our current dataset does not suffer from any of these complications so we’ll simply divide the it in an 80/20 proportion, assigning the larger number of records to the training set.

#divide into test and training sets
#create new col “train” and assign 1 or 0 in 80/20 proportion via random uniform dist
HouseVotes84[,”train”] <- ifelse(runif(nrow(HouseVotes84))<0.80,1,0)
#get col number of train / test indicator column (needed later)
trainColNum <- grep(“train”,names(HouseVotes84))
#separate training and test sets and remove training column before modeling
trainHouseVotes84 <- HouseVotes84[HouseVotes84$train==1,-trainColNum] testHouseVotes84 <- HouseVotes84[HouseVotes84$train==0,-trainColNum]

Now we’re finally good to build our Naive Bayes model (machine learning folks call this model training rather than model building – and I have to admit, it does sound a lot cooler).

The code to train the model is anticlimactically simple:

#load e1071 library and invoke naiveBayes method
library(e1071)
nb_model <- naiveBayes(Class~.,data = trainHouseVotes84)

Here we’ve invokedthe naiveBayes method from the e1071 package. The first argument uses R’s formula notation.In this notation, the dependent variable (to be predicted) appears on the left hand side of the ~ and the independent variables (predictors or features) are on the right hand side. The dot (.) is simply shorthand for “all variable other than the dependent one.” The second argument is the dataframe that contains the training data. Check out the documentation for the other arguments of naiveBayes; it will take me too far afield to cover them here. Incidentally, you can take a look at the model using the summary() or str() functions, or even just entering the model name in the R console:

nb_model
summary(nb_model)
str(nb_model)

Note that I’ve suppressed the output above.

Now that we have a model, we can do some predicting. We do this by feeding our test data into our model and comparing the predicted party affiliations with the known ones. The latter is done via the wonderfully named confusion matrix – a table in which true and predicted values for each of the predicted classes are displayed in a matrix format. This again is just a couple of lines of code:

#…and the moment of reckoning
nb_test_predict <- predict(nb_model,testHouseVotes84[,-1])
#confusion matrix
table(pred=nb_test_predict,true=testHouseVotes84$Class)  pred true democrat republican democrat 38 3 republican 5 22 The numbers you get will be different because your training/test sets are almost certainly different from mine. In the confusion matrix (as defined above), the true values are in columns and the predicted values in rows. So, the algorithm has correctly classified 38 out of 43 (i.e. 38+5) Democrats and 22 out of 25 Republicans (i.e. 22+3). That’s pretty decent. However, we need to keep in mind that this could well be quirk of the choice of dataset. To address this, we should get a numerical measure of the efficacy of the algorithm and for different training and testing datasets. A simple measure of efficacy would be the fraction of predictions that the algorithm gets right. For the training/testing set above, this is simply 60/68 (see the confusion matrix above). The simplest way to calculate this in R is: #fraction of correct predictions mean(nb_test_predict==testHouseVotes84$Class)
[1] 0.8823529

A natural question to ask at this point is: how good is this prediction. This question cannot be answered with only a single run of the model; we need to do many runs and look at the spread of the results. To do this, we’ll create a function which takes the number of times the model should be run and the training fraction as inputs and spits out a vector containing the proportion of correct predictions for each run. Here’s the function

#function to create, run and record model results
nb_multiple_runs <- function(train_fraction,n){
fraction_correct <- rep(NA,n)
for (i in 1:n){
HouseVotes84[,”train”] <- ifelse(runif(nrow(HouseVotes84))<train_fraction,1,0)
trainColNum <- grep(“train”,names(HouseVotes84))
trainHouseVotes84 <- HouseVotes84[HouseVotes84$train==1,-trainColNum] testHouseVotes84 <- HouseVotes84[HouseVotes84$train==0,-trainColNum]
nb_model <- naiveBayes(Class~.,data = trainHouseVotes84)
nb_test_predict <- predict(nb_model,testHouseVotes84[,-1])
fraction_correct[i] <- mean(nb_test_predict==testHouseVotes84$Class) } return(fraction_correct) } I’ve not commented the above code as it is essentially a repeat of the steps described earlier. Also, note that I have not made any effort to make the code generic or efficient. Let’s do 20 runs with the same training fraction (0.8) as before: #20 runs, 80% of data randomly selected for training set in each run fraction_correct_predictions <- nb_multiple_runs(0.8,20) fraction_correct_predictions [1] 0.9417476 0.9036145 0.9294118 0.9302326 0.9213483 0.9404762 0.8777778 0.9102564 [9] 0.9102564 0.9080460 0.9139785 0.9200000 0.9090909 0.9239130 0.9605263 0.9333333 [17] 0.9052632 0.8977273 0.9642857 0.8518519 #summary of results summary(fraction_correct_predictions) Min. 1st Qu. Median Mean 3rd Qu. Max. 0.8519 0.9074 0.9170 0.9177 0.9310 0.9643 #standard deviation sd(fraction_correct_predictions) [1] 0.02582419 We see that the outcome of the runs are quite close together, in the 0.85 to 0.95 range with a standard deviation of 0.025. This tells us that Naive Bayes does a pretty decent job with this data. ### Wrapping up I originally intended to cover a few more case studies in this post, a couple of which highlight the shortcomings of the Naive Bayes algorithm. However, I realize that doing so would make this post unreasonably long, so I’ll stop here with a few closing remarks, and a promise to write up the rest of the story in a subsequent post. To sum up: I have illustrated the use of a popular Naive Bayes implementation in R and attempted to convey an intuition for how the algorithm works. As we have seen, the algorithm works quite well in the example case, despite the violation of the assumption of independent conditional probabilities. The reason for the unreasonable effectiveness of the algorithm is two-fold. Firstly, the algorithm picks the predicted class based on the largest predicted probability, so ordering is more important than the actual value of the probability. Secondly, in many cases, a bias one way for a particular vote may well be counteracted by a bias the other way for another vote. That is, biases tend to cancel out, particularly if there are a large number of features. That said, there are many cases in which the algorithm fails miserably – and we’ll look at some of these in a future post. However, despite its well known shortcomings, Naive Bayes is often the first port of call in prediction problems simply because it is easy to set up and is fast compared to many of the iterative algorithms we will explore later in this series of articles. Endnote Thanks for reading! If you liked this piece, you might enjoy the other articles in my “Gentle introduction to analytics using R” series. Here are the links: A gentle introduction to text mining using R A gentle introduction to cluster analysis using R A gentle introduction to topic modeling using R Written by K November 6, 2015 at 7:33 am ## A gentle introduction to topic modeling using R with 45 comments ### Introduction The standard way to search for documents on the internet is via keywords or keyphrases. This is pretty much what Google and other search engines do routinely…and they do it well. However, as useful as this is, it has its limitations. Consider, for example, a situation in which you are confronted with a large collection of documents but have no idea what they are about. One of the first things you might want to do is to classify these documents into topics or themes. Among other things this would help you figure out if there’s anything interest while also directing you to the relevant subset(s) of the corpus. For small collections, one could do this by simply going through each document but this is clearly infeasible for corpuses containing thousands of documents. Topic modeling – the theme of this post – deals with the problem of automatically classifying sets of documents into themes The article is organised as follows: I first provide some background on topic modelling. The algorithm that I use, Latent Dirichlet Allocation (LDA), involves some pretty heavy maths which I’ll avoid altogether. However, I will provide an intuitive explanation of how LDA works before moving on to a practical example which uses the topicmodels library in R. As in my previous articles in this series (see this post and this one), I will discuss the steps in detail along with explanations and provide accessible references for concepts that cannot be covered in the space of a blog post. (Aside: Beware, LDA is also an abbreviation for Linear Discriminant Analysis a classification technique that I hope to cover later in my ongoing series on text and data analytics). ### Latent Dirichlet Allocation – a math-free introduction In essence, LDA is a technique that facilitates the automatic discovery of themes in a collection of documents. The basic assumption behind LDA is that each of the documents in a collection consist of a mixture of collection-wide topics. However, in reality we observe only documents and words, not topics – the latter are part of the hidden (or latent) structure of documents. The aim is to infer the latent topic structure given the words and document. LDA does this by recreating the documents in the corpus by adjusting the relative importance of topics in documents and words in topics iteratively. Here’s a brief explanation of how the algorithm works, quoted directly from this answer by Edwin Chen on Quora: • Go through each document, and randomly assign each word in the document to one of the K topics. (Note: One of the shortcomings of LDA is that one has to specify the number of topics, denoted by K, upfront. More about this later.) • This assignment already gives you both topic representations of all the documents and word distributions of all the topics (albeit not very good ones). • So to improve on them, for each document d… • ….Go through each word w in d… • ……..And for each topic t, compute two things: 1) p(topic t | document d) = the proportion of words in document d that are currently assigned to topic t, and 2) p(word w | topic t) = the proportion of assignments to topic t over all documents that come from this word w. Reassign w a new topic, where you choose topic t with probability p(topic t | document d) * p(word w | topic t) (according to our generative model, this is essentially the probability that topic t generated word w, so it makes sense that we resample the current word’s topic with this probability). (Note: p(a|b) is the conditional probability of a given that b has already occurred – see this post for more on conditional probabilities) • ……..In other words, in this step, we’re assuming that all topic assignments except for the current word in question are correct, and then updating the assignment of the current word using our model of how documents are generated. • After repeating the previous step a large number of times, you’ll eventually reach a roughly steady state where your assignments are pretty good. So use these assignments to estimate the topic mixtures of each document (by counting the proportion of words assigned to each topic within that document) and the words associated to each topic (by counting the proportion of words assigned to each topic overall). For another simple explanation of how LDA works in, check out this article by Matthew Jockers. For a more technical exposition, take a look at this video by David Blei, one of the inventors of the algorithm. The iterative process described in the last point above is implemented using a technique called Gibbs sampling. I’ll say a bit more about Gibbs sampling later, but you may want to have a look at this paper by Philip Resnick and Eric Hardesty that explains the nitty-gritty of the algorithm (Warning: it involves a fair bit of math, but has some good intuitive explanations as well). As a general point, I should also emphasise that you do not need to understand the ins and outs of an algorithm to use it but it does help to understand, at least at a high level, what the algorithm is doing. One needs to develop a feel for algorithms even if one doesn’t understand the details. Indeed, most people working in analytics do not know the details of the algorithms they use, but that doesn’t stop them from using algorithms intelligently. Purists may disagree. I think they are wrong. Finally – because you’re no doubt wondering 🙂 – the term “Dirichlet” in LDA refers to the fact that topics and words are assumed to follow Dirichlet distributions. There is no “good” reason for this apart from convenience – Dirichlet distributions provide good approximations to word distributions in documents and, perhaps more important, are computationally convenient. ### Preprocessing As in my previous articles on text mining, I will use a collection of 30 posts from this blog as an example corpus. The corpus can be downloaded here. I will assume that you have R and RStudio installed. Follow this link if you need help with that. The preprocessing steps are much the same as described in my previous articles. Nevertheless, I’ll risk boring you with a detailed listing so that you can reproduce my results yourself: #load text mining library library(tm) #set working directory (modify path as needed) setwd(“C:\\Users\\Kailash\\Documents\\TextMining”) #load files into corpus #get listing of .txt files in directory filenames <- list.files(getwd(),pattern=”*.txt”) #read files into a character vector files <- lapply(filenames,readLines) #create corpus from vector docs <- Corpus(VectorSource(files)) #inspect a particular document in corpus writeLines(as.character(docs[[30]])) #start preprocessing #Transform to lower case docs <-tm_map(docs,content_transformer(tolower)) #remove potentially problematic symbols toSpace <- content_transformer(function(x, pattern) { return (gsub(pattern, ” “, x))}) docs <- tm_map(docs, toSpace, “-“) docs <- tm_map(docs, toSpace, “’”) docs <- tm_map(docs, toSpace, “‘”) docs <- tm_map(docs, toSpace, “•”) docs <- tm_map(docs, toSpace, “””) docs <- tm_map(docs, toSpace, ““”) #remove punctuation docs <- tm_map(docs, removePunctuation) #Strip digits docs <- tm_map(docs, removeNumbers) #remove stopwords docs <- tm_map(docs, removeWords, stopwords(“english”)) #remove whitespace docs <- tm_map(docs, stripWhitespace) #Good practice to check every now and then writeLines(as.character(docs[[30]])) #Stem document docs <- tm_map(docs,stemDocument) #fix up 1) differences between us and aussie english 2) general errors docs <- tm_map(docs, content_transformer(gsub), pattern = “organiz”, replacement = “organ”) docs <- tm_map(docs, content_transformer(gsub), pattern = “organis”, replacement = “organ”) docs <- tm_map(docs, content_transformer(gsub), pattern = “andgovern”, replacement = “govern”) docs <- tm_map(docs, content_transformer(gsub), pattern = “inenterpris”, replacement = “enterpris”) docs <- tm_map(docs, content_transformer(gsub), pattern = “team-“, replacement = “team”) #define and eliminate all custom stopwords myStopwords <- c(“can”, “say”,”one”,”way”,”use”, “also”,”howev”,”tell”,”will”, “much”,”need”,”take”,”tend”,”even”, “like”,”particular”,”rather”,”said”, “get”,”well”,”make”,”ask”,”come”,”end”, “first”,”two”,”help”,”often”,”may”, “might”,”see”,”someth”,”thing”,”point”, “post”,”look”,”right”,”now”,”think”,”‘ve “, “‘re “,”anoth”,”put”,”set”,”new”,”good”, “want”,”sure”,”kind”,”larg”,”yes,”,”day”,”etc”, “quit”,”sinc”,”attempt”,”lack”,”seen”,”awar”, “littl”,”ever”,”moreov”,”though”,”found”,”abl”, “enough”,”far”,”earli”,”away”,”achiev”,”draw”, “last”,”never”,”brief”,”bit”,”entir”,”brief”, “great”,”lot”) docs <- tm_map(docs, removeWords, myStopwords) #inspect a document as a check writeLines(as.character(docs[[30]])) #Create document-term matrix dtm <- DocumentTermMatrix(docs) #convert rownames to filenames rownames(dtm) <- filenames #collapse matrix by summing over columns freq <- colSums(as.matrix(dtm)) #length should be total number of terms length(freq) #create sort order (descending) ord <- order(freq,decreasing=TRUE) #List all terms in decreasing order of freq and write to disk freq[ord] write.csv(freq[ord],”word_freq.csv”) Check out the preprocessing section in either this article or this one for detailed explanations of the code. The document term matrix (DTM) produced by the above code will be the main input into the LDA algorithm of the next section. ### Topic modelling using LDA We are now ready to do some topic modelling. We’ll use the topicmodels package written by Bettina Gruen and Kurt Hornik. Specifically, we’ll use the LDA function with the Gibbs sampling option mentioned earlier, and I’ll say more about it in a second. The LDA function has a fairly large number of parameters. I’ll describe these briefly below. For more, please check out this vignette by Gruen and Hornik. For the most part, we’ll use the default parameter values supplied by the LDA function,custom setting only the parameters that are required by the Gibbs sampling algorithm. Gibbs sampling works by performing a random walk in such a way that reflects the characteristics of a desired distribution. Because the starting point of the walk is chosen at random, it is necessary to discard the first few steps of the walk (as these do not correctly reflect the properties of distribution). This is referred to as the burn-in period. We set the burn-in parameter to 4000. Following the burn-in period, we perform 2000 iterations, taking every 500th iteration for further use. The reason we do this is to avoid correlations between samples. We use 5 different starting points (nstart=5) – that is, five independent runs. Each starting point requires a seed integer (this also ensures reproducibility), so I have provided 5 random integers in my seed list. Finally I’ve set best to TRUE (actually a default setting), which instructs the algorithm to return results of the run with the highest posterior probability. Some words of caution are in order here. It should be emphasised that the settings above do not guarantee the convergence of the algorithm to a globally optimal solution. Indeed, Gibbs sampling will, at best, find only a locally optimal solution, and even this is hard to prove mathematically in specific practical problems such as the one we are dealing with here. The upshot of this is that it is best to do lots of runs with different settings of parameters to check the stability of your results. The bottom line is that our interest is purely practical so it is good enough if the results make sense. We’ll leave issues of mathematical rigour to those better qualified to deal with them 🙂 As mentioned earlier, there is an important parameter that must be specified upfront: k, the number of topics that the algorithm should use to classify documents. There are mathematical approaches to this, but they often do not yield semantically meaningful choices of k (see this post on stackoverflow for an example). From a practical point of view, one can simply run the algorithm for different values of k and make a choice based by inspecting the results. This is what we’ll do. OK, so the first step is to set these parameters in R… and while we’re at it, let’s also load the topicmodels library (Note: you might need to install this package as it is not a part of the base R installation). #load topic models library library(topicmodels) #Set parameters for Gibbs sampling burnin <- 4000 iter <- 2000 thin <- 500 seed <-list(2003,5,63,100001,765) nstart <- 5 best <- TRUE #Number of topics k <- 5 That done, we can now do the actual work – run the topic modelling algorithm on our corpus. Here is the code: #Run LDA using Gibbs sampling ldaOut <-LDA(dtm,k, method=”Gibbs”, control=list(nstart=nstart, seed = seed, best=best, burnin = burnin, iter = iter, thin=thin)) #write out results #docs to topics ldaOut.topics <- as.matrix(topics(ldaOut)) write.csv(ldaOut.topics,file=paste(“LDAGibbs”,k,”DocsToTopics.csv”)) #top 6 terms in each topic ldaOut.terms <- as.matrix(terms(ldaOut,6)) write.csv(ldaOut.terms,file=paste(“LDAGibbs”,k,”TopicsToTerms.csv”)) #probabilities associated with each topic assignment topicProbabilities <- as.data.frame(ldaOut@gamma) write.csv(topicProbabilities,file=paste(“LDAGibbs”,k,”TopicProbabilities.csv”)) #Find relative importance of top 2 topics topic1ToTopic2 <- lapply(1:nrow(dtm),function(x) sort(topicProbabilities[x,])[k]/sort(topicProbabilities[x,])[k-1]) #Find relative importance of second and third most important topics topic2ToTopic3 <- lapply(1:nrow(dtm),function(x) sort(topicProbabilities[x,])[k-1]/sort(topicProbabilities[x,])[k-2]) #write to file write.csv(topic1ToTopic2,file=paste(“LDAGibbs”,k,”Topic1ToTopic2.csv”)) write.csv(topic2ToTopic3,file=paste(“LDAGibbs”,k,”Topic2ToTopic3.csv”)) The LDA algorithm returns an object that contains a lot of information. Of particular interest to us are the document to topic assignments, the top terms in each topic and the probabilities associated with each of those terms. These are printed out in the first three calls to write.csv above. There are a few important points to note here: 1. Each document is considered to be a mixture of all topics (5 in this case). The assignments in the first file list the top topic – that is, the one with the highest probability (more about this in point 3 below). 2. Each topic contains all terms (words) in the corpus, albeit with different probabilities. We list only the top 6 terms in the second file. 3. The last file lists the probabilities with which each topic is assigned to a document. This is therefore a 30 x 5 matrix – 30 docs and 5 topics. As one might expect, the highest probability in each row corresponds to the topic assigned to that document. The “goodness” of the primary assignment (as discussed in point 1) can be assessed by taking the ratio of the highest to second-highest probability and the second-highest to the third-highest probability and so on. This is what I’ve done in the last nine lines of the code above. Take some time to examine the output and confirm for yourself that that the primary topic assignments are best when the ratios of probabilities discussed in point 3 are highest. You should also experiment with different values of k to see if you can find better topic distributions. In the interests of space I will restrict myself to k = 5. The table below lists the top 6 terms in topics 1 through 5.  Topic 1 Topic 2 Topic 3 Topic 4 Topic 5 1 work question chang system project 2 practic map organ data manag 3 mani time consult model approach 4 flexibl ibi manag design organ 5 differ issu work process decis 6 best plan problem busi problem The table below lists the document to (primary) topic assignments:  Document Topic BeyondEntitiesAndRelationships.txt 4 bigdata.txt 4 ConditionsOverCauses.txt 5 EmergentDesignInEnterpriseIT.txt 4 FromInformationToKnowledge.txt 2 FromTheCoalface.txt 1 HeraclitusAndParmenides.txt 3 IroniesOfEnterpriseIT.txt 3 MakingSenseOfOrganizationalChange.txt 5 MakingSenseOfSensemaking.txt 2 ObjectivityAndTheEthicalDimensionOfDecisionMaking.txt 5 OnTheInherentAmbiguitiesOfManagingProjects.txt 5 OrganisationalSurprise.txt 5 ProfessionalsOrPoliticians.txt 3 RitualsInInformationSystemDesign.txt 4 RoutinesAndReality.txt 4 ScapegoatsAndSystems.txt 5 SherlockHolmesFailedProjects.txt 3 sherlockHolmesMgmtFetis.txt 3 SixHeresiesForBI.txt 4 SixHeresiesForEnterpriseArchitecture.txt 3 TheArchitectAndTheApparition.txt 3 TheCloudAndTheGrass.txt 2 TheConsultantsDilemma.txt 3 TheDangerWithin.txt 5 TheDilemmasOfEnterpriseIT.txt 3 TheEssenceOfEntrepreneurship.txt 1 ThreeTypesOfUncertainty.txt 5 TOGAFOrNotTOGAF.txt 3 UnderstandingFlexibility.txt 1 From a quick perusal of the two tables it appears that the algorithm has done a pretty decent job. For example,topic 4 is about data and system design, and the documents assigned to it are on topic. However, it is far from perfect – for example, the interview I did with Neil Preston on organisational change (MakingSenseOfOrganizationalChange.txt) has been assigned to topic 5, which seems to be about project management. It ought to be associated with Topic 3, which is about change. Let’s see if we can resolve this by looking at probabilities associated with topics. The table below lists the topic probabilities by document:  Topic 1 Topic 2 Topic 3 Topic 4 Topic 5 BeyondEn 0.071 0.064 0.024 0.741 0.1 bigdata. 0.182 0.221 0.182 0.26 0.156 Conditio 0.144 0.109 0.048 0.205 0.494 Emergent 0.121 0.226 0.204 0.236 0.213 FromInfo 0.096 0.643 0.026 0.169 0.066 FromTheC 0.636 0.082 0.058 0.086 0.138 Heraclit 0.137 0.091 0.503 0.162 0.107 IroniesO 0.101 0.088 0.388 0.26 0.162 MakingSe 0.13 0.206 0.262 0.089 0.313 MakingSe 0.09 0.715 0.055 0.067 0.074 Objectiv 0.216 0.078 0.086 0.242 0.378 OnTheInh 0.18 0.234 0.102 0.12 0.364 Organisa 0.089 0.095 0.07 0.092 0.655 Professi 0.155 0.064 0.509 0.128 0.144 RitualsI 0.103 0.064 0.044 0.676 0.112 Routines 0.108 0.042 0.033 0.69 0.127 Scapegoa 0.135 0.088 0.043 0.185 0.549 Sherlock 0.093 0.082 0.398 0.195 0.232 sherlock 0.108 0.136 0.453 0.123 0.18 SixHeres 0.159 0.11 0.078 0.516 0.138 SixHeres 0.104 0.111 0.366 0.212 0.207 TheArchi 0.111 0.221 0.522 0.088 0.058 TheCloud 0.185 0.333 0.198 0.136 0.148 TheConsu 0.105 0.184 0.518 0.096 0.096 TheDange 0.114 0.079 0.037 0.079 0.69 TheDilem 0.125 0.128 0.389 0.261 0.098 TheEssen 0.713 0.059 0.031 0.113 0.084 ThreeTyp 0.09 0.076 0.042 0.083 0.708 TOGAFOrN 0.158 0.232 0.352 0.151 0.107 Understa 0.658 0.065 0.072 0.101 0.105 In the table, the highest probability in each row is in bold. Also, in cases where the maximum and the second/third largest probabilities are close, I have highlighted the second (and third) highest probabilities in red. It is clear that Neil’s interview (9th document in the above table) has 3 topics with comparable probabilities – topic 5 (project management), topic 3 (change) and topic 2 (issue mapping / ibis), in decreasing order of probabilities. In general, if a document has multiple topics with comparable probabilities, it simply means that the document speaks to all those topics in proportions indicated by the probabilities. A reading of Neil’s interview will convince you that our conversation did indeed range over all those topics. That said, the algorithm is far from perfect. You might have already noticed a few poor assignments. Here is one – my post on Sherlock Holmes and the case of the failed project has been assigned to topic 3; I reckon it belongs in topic 5. There are a number of others, but I won’t belabor the point, except to reiterate that this precisely why you definitely want to experiment with different settings of the iteration parameters (to check for stability) and, more important, try a range of different values of k to find the optimal number of topics. ### To conclude Topic modelling provides a quick and convenient way to perform unsupervised classification of a corpus of documents. As always, though, one needs to examine the results carefully to check that they make sense. I’d like to end with a general observation. Classifying documents is an age-old concern that cuts across disciplines. So it is no surprise that topic modelling has got a look-in from diverse communities. Indeed, when I was reading up and learning about LDA, I found that some of the best introductory articles in the area have been written by academics working in English departments! This is one of the things I love about working in text analysis, there is a wealth of material on the web written from diverse perspectives. The term cross-disciplinary often tends to be a platitude , but in this case it is simply a statement of fact. I hope that I have been able to convince you to explore this rapidly evolving field. Exciting times ahead, come join the fun. Written by K September 29, 2015 at 7:18 pm ## Setting up an internal data analytics practice – some thoughts from a wayfarer leave a comment » ### Introduction This year has been hugely exciting so far: I’ve been exploring and playing with various techniques that fall under the general categories of data mining and text analytics. What’s been particularly satisfying is that I’ve been fortunate to find meaningful applications for these techniques within my organization. Although I have a fair way to travel yet, I’ve learnt that common wisdom about data analytics – especially the stuff that comes from software vendors and high-end consultancies – can be misleading, even plain wrong. Hence this post in which I dispatch some myths and share a few pointers on establishing data analytics capabilities within an organization. ### Busting a few myths Let’s get right to it by taking a critical look at a few myths about setting up an internal data analytics practice. 1. Requires high-end technology and a big budget: this myth is easy to bust because I can speak from recent experience. No, you do not need cutting-edge technology or an oversized budget. You can get started for with an outlay of 0$ – yes, that’s right, for free!  All you need to is the open-source statistical package R (check out this section of my article on text mining for more on installing and using R) and the willingness to roll-up your sleeves and learn (more about this  later).  No worries if you prefer to stick with familiar tools – you can even begin with Excel.
2. Needs specialist skills: another myth floating around is that you need Phd level knowledge in statistics or applied mathematics to do practical work in analytics. Sorry, but that’s plain wrong. You do need a PhD to do research in the analytics and develop your own algorithms, but not if you want to apply algorithms written by others.Yes, you will need to develop an understanding of the algorithms you plan to use, a feel for how they work and the ability to tell whether the results make sense. There are many good resources that can help you develop these skills – see, for example, the outstanding books by James, Witten, Hastie and Tibshirani and Kuhn and Johnson.
3. Must have sponsorship from the top: this one is possibly a little more controversial than the previous two. It could be argued that it is impossible to gain buy in for a new capability without sponsorship from top management. However, in my experience, it is OK to start small by finding potential internal “customers” for analytics services through informal conversations with folks in different functions.I started by having informal conversations with managers in two different areas: IT infrastructure and sales / marketing.  I picked these two areas because I knew that they had several gigabytes of under-exploited data – a good bit of it unstructured – and a lot of open questions that could potentially be answered (at least partially) via methods of data and text analytics.  It turned out I was right. I’m currently doing a number of proofs of concept and small projects in both these areas.  So you don’t need sponsorship from the top as long as you can get buy in from people who have problems they believe you can solve. If you deliver, they may even advocate your cause to their managers.

A caveat is in order at this point:  my organization is not the same as yours, so you may well need to follow a different path from mine. Nevertheless, I do believe that it is always possible to find a way to start without needing permission or incurring official wrath.  In that spirit, I now offer some suggestions to help kick-start your efforts

### Getting started

As the truism goes, the hardest part of any new effort is getting started.  The first thing to keep in mind is to start small. This is true even if you have official sponsorship and a king-sized budget. It is very tempting to spend a lot of time garnering management support for investing in high-end technology.  Don’t do it!  Do the following instead:

1. Develop an understanding of the problems faced by people you plan to approach: The best way to do this is to talk to analysts or frontline managers. In my case, I was fortunate to have access to some very savvy analysts in IT service management and marketing who gave me a slew of interesting ideas to pursue. A word of advice: it is best not to talk to senior managers until you have a few concrete results that you can quantify in terms of dollar values.
2. Invest time and effort in understanding analytics algorithms and gaining practical experience with them: As mentioned earlier, I started with R – and I believe it is the best choice. Not just because it is free but also because there are a host of packages available to tackle just about any analytics problem you might encounter.  There are some excellent free resources available to get you started with R (check out this listing on the r-statistics blog, for example).It is important that you start cutting code as you learn. This will help you build a repertoire of techniques and approaches as you progress. If you get stuck when coding, chances are you will find a solution on the wonderful stackoverflow site.
3. Evangelise, evangelise, evangelise: You are, in effect, trying to sell an idea to people within your organization. You therefore have to identify people who might be able to help you and then convince them that your idea has merit. The best way to do the latter is to have concrete examples of problems that you have tackled. This is a chicken-and-egg situation in that you can’t have any examples until you gain support.  I got support by approaching people I know well. I found that most – no, all – of them were happy to provide me with interesting ideas and access to their data.
4. Begin with small (but real) problems: It is important to start with the “low-hanging fruit” – the problems that would take the least effort to solve. However, it is equally important to address real problems, i.e. those that matter to someone.
5. Leverage your organisation’s existing data infrastructure: From what I’ve written thus far, I may have given you the impression that the tools of data analytics stand separate from your existing data infrastructure. Nothing could be further from the truth. In reality, I often do the initial work  (basic preprocessing and exploratory analysis) using my organisation’s relational database infrastructure. Relational databases have sophisticated analytical extensions to SQL as well as efficient bulk data cleansing and transport facilities. Using these make good sense, particularly if your R installation is on a desktop or laptop computer as it is in my case. Moreover, many enterprise database vendors now offer add-on options that integrate R with their products. This gives you the best of both worlds – relational and analytical capabilities on an enterprise-class platform.
6. Build relationships with the data management team: Remember the work you are doing falls under the ambit of the group that is officially responsible for managing data in your organization. It is therefore important that you keep them informed of what you’re doing.  Sooner or later your paths will cross, and you want to be sure that there are no nasty surprises (for either side!) at that point. Moreover, if you build connections with them early, you may even find that the data management team supports your efforts.

Having waxed verbose, I should mention that my effort is work in progress and I do not know where it will lead. Nevertheless, I offer these suggestions as a wayfarer who is considerably further down the road from where he started.

### Parting thoughts

You may have noticed that I’ve refrained from using the overused and over-hyped term “Big Data” in this piece. This is deliberate. Indeed, the techniques I have been using have nothing to do with the size of the datasets. To be honest, I’ve applied them to datasets ranging from a few thousand to a few hundred thousand records, both of which qualify as Very Small Data in today’s world.

Your vendor will be only too happy to sell you Big Data infrastructure that will set you back a good many dollars. However, the chances are good that you do not need it right now.  You’ll be much better off going back to them after you hit the limits of your current data processing infrastructure. Moreover, you’ll also be better informed about your needs then.

You may also be wondering why I haven’t said much about the composition of the analytics team (barring the point about not needing PhD statisticians) and how it should be organized.  The reason I haven’t done so is that I believe the right composition and organizational structure will emerge from the initial projects done and feedback received from internal customers. The resulting structure will be better suited to the organization than one that is imposed upfront.  Long time readers of this blog might recognize this as a tenet of emergent design.

Finally, I should reiterate that my efforts are still very much in progress and I know not where they will lead. However, even if they go nowhere, I would have learnt something about my organization and picked up a useful, practical skill. And that is good enough for me.

Written by K

September 3, 2015 at 8:28 pm

Tagged with