Many applications require the selection of a subset of objects from a larger set. For example, suppose you are down-sampling a large data set by choosing the indices that you want to keep at random. If you are satisfied with obtaining duplicate indices (in essence, if you are sampling with replacement), this is a trivial matter in most programming language. Python certainly makes this absurdly easy:

import random

def sample(k,n):
return [random.choice( [x for x in range(n)] ) for _ in range(k)]


But what about sampling without replacement? More precisely, what if you require the selection of k indices from an array of n values without choosing duplicate values but still want the probability of choosing a certain element to be uniform? Well, if you use Python, you are in luck again:

import random

def sample_without_replacement(k,n):
return random.sample([x for x in range(n)], k)


But what about other programming languages that may not have similar builtin functionality, such as C++? In this case, a very simple and elegant technique is given by none other than the great Donald E. Knuth. The following algorithm originally was given in The Art of Computer Programming, Volume 2, Section 3.4.2 on p. 137. Briefly put, it works like this:

1. Set t = 0 and m = 0. m represents the number of items selected so far, while t is the total number of items that have already been traversed.

2. Generate a random number U that is uniformly distributed between zero and one.

3. If (N-t)*U >= n-m, skip the current item by increasing t by 1 and going back to step 2. Else, select the current item by increasing both t and m by 1. Afterwards, either go back to step 2 or stop the algorithm (if sufficiently many items have been sampled already).

A very basic implementation of this algorithm (in C++) is given in the gist below:

I do not know about you, but my gut reaction upon seeing this algorithm for the first time was How can this possibly work? So, let us briefly prove the algorithm to be correct. Basically, we have to show that every item has the same probability of being chosen. The key insight is to realize that the probability has to change depending on the number of elements that have already been selected. More precisely, we need to determine the probability of choosing item t+1 if m items have already been selected. This requires some combinatorics. There are ways of choosing n items from N such that m items are picked as the first t items. Or, equivalently, these are the number of potential permutations of the remaining n-m elements. Out of these, we are interested in all the ones that contain item t+1—but this is easy, because we can just take that item as granted and count the remaining combinations as , i.e. the number of ways to choose n-m-1 items out of N-t-1 ones. The quotient of these two numbers is exactly the probability with which we must choose item t+1 if we want to have a uniform probability for choosing a certain item. This quotient turns out to be , which looks familiar. Finally, note that since U was chosen to be uniformly distributed between zero and one, the condition (N-t)*U >= n-m in the algorithm is satisfied with the required probability. Consequently, this method samples without replacement in the required manner!

If you are interested or require a more efficient algorithm, please refer to the paper An Efficient Algorithm for Sequential Random Sampling by Jeffrey Scott Vitter in ACM Transactions on Mathematical Software, Volume 13, Issue 1, pp. 58–67, for more details. The paper should be available free-of-charge from the link provided above, but there is also a mirror available, thanks to the good folks at INRIA, who require their publications to be kept in an open archive. Time permitting, I might provide a second blog post and implementation about the more efficient algorithm.

That is all for now, until next time!

Posted late Sunday evening, October 8th, 2017 Tags:

A warning upfront: this post is sort of an advertisement for Aleph, my C++ library for topological data analysis. In this blog post I do not want to cover any of the mathematical algorithms present in Aleph—I rather want to focus on small but integral part of the project, viz. unit tests.

If you are not familiar with the concept of unit testing, the idea is (roughly) to write a small, self-sufficient test for every new piece of functionality that you write. Proponents of the methodology of test-driven development (TDD) even go so far as to require you to write the unit tests before you write your actual code. In this mindset, you first think of the results your code should achieve and which outputs you expect prior to writing any “real” code. I am putting the word real in quotation marks here because it may seem strange to focus on the tests before doing the heavy lifting.

However, this way of approaching software development may actually be quite beneficial, in particular if you are working on algorithms with a nice mathematical flavour. Here, thinking about the results you want to achieve with your code ensures that at least a few known examples are processed correctly by your code, making it more probable that the code will perform well in real-world scenarios.

When I started writing Aleph in 2016, I also wanted to add some unit tests, but I did not think that the size of the library warranted the inclusion of one of the big players, such as Google Test or Boost.Test. While arguably extremely powerful and teeming with more features than I could possibly imagine, they are also quite heavy and require non-trivial adjustments to any project.

Thus, in the best tradition of the not-invented-here-syndrome, I decided to roll my own testing framework, base on pure CMake and small dash of C++. My design decisions were rather simple:

• Use CTest, the testing framework of CMake to run the tests. This framework is rather simple and just uses the return type of a unit test program to decide whether the test worked correctly.
• Provide a set of routines to check the correctness of certain calculations within a unit test, throwing an error if something unexpected happened.
• Collect unit tests for the “larger” parts of the project in a single executable program.

Yes, you read that right—my approach actually opts for throwing an error in order to crash the unit test program. Bear with me, though, for I think that this is actually a rather sane way of approaching unit tests. After all, if the tests fails, I am usually not interested in whether other parts of a test program—that may potentially depend on previous calculations—run through or not. As a consequence, adding a unit test to Aleph is as simple as adding the following lines to a CMakeLists.txt file, located in the tests subdirectory of the project:

ADD_EXECUTABLE( test_io_gml test_io_gml.cc )


While in the main CMakeLists.txt, I added the following lines:

ENABLE_TESTING()


So far, so good. A test now looks like this:

#include <tests/Base.hh>

void testBasic()
{
// do some nice calculation; store the results in foo and bar,
// respectively

ALEPH_ASSERT_THROW( foo != bar );
ALEPH_ASSERT_EQUAL( foo, 2.0 );
ALEPH_ASSERT_EQUAL( bar, 1.0 );
}

{
}

int main(int, char**)
{
testBasic();
}


That is basically the whole recipe for a simple unit test. Upon execution, main() will ensure that all larger-scale test routines, i.e. testSimple() and testAdvanced() are called. Within each of these routines, the calls to the corresponding macros—more on that in a minute— ensure that conditions are met, or certain values are equal to other values. Else, an error will be thrown, the test will abort, and CMake will throw an error upon test execution.

So, how do the macros look like? Here is a copy of the current version of Aleph:

#define ALEPH_ASSERT_THROW( condition )                             \
{                                                                   \
if( !( condition ) )                                              \
{                                                                 \
throw std::runtime_error(   std::string( __FILE__ )             \
+ std::string( ":" )                  \
+ std::to_string( __LINE__ )          \
+ std::string( " in " )               \
+ std::string( __PRETTY_FUNCTION__ )  \
);                                                              \
}                                                                 \
}

#define ALEPH_ASSERT_EQUAL( x, y )                                  \
{                                                                   \
if( ( x ) != ( y ) )                                              \
{                                                                 \
throw std::runtime_error(   std::string( __FILE__ )             \
+ std::string( ":" )                  \
+ std::to_string( __LINE__ )          \
+ std::string( " in " )               \
+ std::string( __PRETTY_FUNCTION__ )  \
+ std::string( ": " )                 \
+ std::to_string( ( x ) )             \
+ std::string( " != " )               \
+ std::to_string( ( y ) )             \
);                                                              \
}                                                                 \
}


Pretty simple, I would say. The ALEPH_ASSERT_EQUAL macro actually tries to convert the corresponding values to strings, which may not always work. Of course, you could use more complicated string conversion routines, as Boost.Test does. For now, though, these macros are sufficient to make up the unit test framework of Aleph, which at the time of me writing this, encompasses more than 4000 lines of code.

The only remaining question is how this framework is used in practice. By setting ENABLE_TESTING(), CMake actually exposes a new target called test. Hence, in order to run those tests, a simple make test is sufficient in the build directory. This is what the result may look like:

$make test Running tests... Test project /home/bastian/Projects/Aleph/build Start 1: barycentric_subdivision 1/36 Test #1: barycentric_subdivision ............ Passed 0.00 sec Start 2: beta_skeleton [...] 34/36 Test #34: union_find ......................... Passed 0.00 sec Start 35: witness_complex 35/36 Test #35: witness_complex .................... Passed 1.82 sec Start 36: python_integration 36/36 Test #36: python_integration ................. Passed 0.07 sec 100% tests passed, 0 tests failed out of 36 Total Test time (real) = 5.74 sec  In addition to being rather lean, this framework can easily be integrated into an existing Travis CI workflow by adding - make test  as an additional step to the script target in your .travis.yml file. If you are interested in using this testing framework, please take a look at the following files: That is all for now, until next time—may your unit tests always work the way you expect them to! Posted Sunday night, October 15th, 2017 Software developers are already familiar with the KISS principle. Roughly speaking, it refers to the old wisdom that simple solutions are to be preferred, while unnecessary complexity should be avoided. In machine learning, this means that one should increase the complexity iteratively, starting from simple models and—if necessary—working upwards to more complicated ones. I was recently reminded and humbled by the wisdom of KISS while working on a novel topology-based kernel for machine learning. # Prelude: A Paper on Deep Graph Kernels A recent paper by Yanardag and Vishwanathan introduced a way of combining graph kernels with deep learning techniques. The new framework basically yields a way of using graph kernels, such as graphlet kernels in a regular deep learning setting. The two researchers used some interesting data sets for their publication. Most notably, they extracted a set of co-occurrence networks (more about that in a minute) from Reddit, a content aggregation and discussion site. Reddit consists of different communities, the subreddits. Each subreddit deals with a different topic, ranging from archaeology to zoology. The posting style of these subreddits varies a lot. There are several subreddits that are based on a question–answer format, while others are more centred around individual discussions. Yanardag and Vishwanathan hence crawled the top submissions from the subreddits IamA, AskReddit, both of which are based on questions and answers, as well as from TrollXChromosomes, and atheism, which are discussion-based subreddits. From every submission, a graph was created by taking all the commenters of a thread as nodes and connecting two nodes by an edge if one user responds to the comment of another user. We can see that this is an extremely simple model—it represents only a fraction of the information available in every discussion. Nonetheless, there is some hope that qualitatively different behaviours will emerge. More precisely, the assumption of Yanardag and Vishwanathan is that there is some quantifiable difference between question–answer subreddits and discussion-based subreddits. Their paper aims to learn the correct classification for each thread. Hence, given a graph, we want to teach the computer to tell us whether the graph is more likely to arise from a question–answer subreddit or from a discussion-based one. The two researchers refer to this data set as REDDIT-BINARY. It is now available in a repository of benchmark data sets for graph kernels, gracefully provided and lovingly curated by the CS department of Dortmund University. # Interlude: Looking at the Data Prior to actually working with the data, let us first take a look at it in order to get a feel for its inherent patterns. To this end, I used Aleph, my library for topological data analysis read and convert the individual graphs of every discussion to the simpler GML format. See below if you are also interested in the data. Next, I used Gephi to obtain a force-directed visualization of some of the networks. A look at a question–answer subreddit shows a central core structure of the data set, from which numerous strands—each corresponding most probably to smaller discussions—emerge: The discussion-based subreddit graph, on the other hand, exhibits a larger depth, manifesting themselves in a few long strands. Nonetheless, a similar central core structure is observable as well. Keep in mind that the selected examples are not necessarily representative—I merely picked two of the large graphs in the data set to obtain an idea of how the data looks. # A Complex Classification A straightforward way to classify those networks would be to use graph kernels, such as the graphlet kernels. The basic idea behind these kernels is to measure a dissimilarity between two graphs by means of, for example, the presence or absence of certain subgraphs. At least this is the strategy pursued by the graphlet kernel. Other kernels may instead opt for comparing random walks on both graphs. A common theme of these kernels is that they are rather expensive to compute. In many applications, they are the only hope of obtaining suitable dissimilarity information without having to solve the graph isomorphism problem, which is even more computationally expensive. Hence, graph kernels are often the only suitable way of assessing the dissimilarity between graphs. They are quite flexible and, as the authors show, can even be integrated into a deep learning framework. As for their performance, Yanardag and Vishwanathan report an accuracy of 77.34% (standard deviation 0.18) for graphlet kernels and 78.04% (standard deviation 0.39) for deep graphlet kernels, which is very good considering the baseline for random guessing is 50%. # A Simple Classification In line with the KISS principle, I wanted to figure out alternative ways of achieving similar accuracy values with a better performance, if possible. This suggests a feature-based approach with features that can be calculated easily. There are numerous potential options for choosing these features. I figured that good choices include the average clustering coefficient of the graph, the average degree, the average shortest path length, the density, and the diameter of the graph. Among these, the average shortest path length and the diameter take longest to compute because they essentially have to enumerate all shortest paths in the graph. So I only used the remaining three features, all of which are computable in polynomial time. I used the excellent NetworkX package for Python to do most of the heavy lifting. Reading a graph and calculating its features is as easy as it gets: import networkx as nx G = nx.read_gml(filename, label='id') average_clustering_coefficient = nx.average_clustering(G) average_degree = np.mean( [degree for _,degree in nx.degree(G) ] ) density = nx.density(G)  I collect these values in a pandas.DataFrame to simplify their handling. Now we have to choose some classifiers. I selected decision trees, support vector machines, and logistic regression. Next, let us train and test each of these classifiers by means of 10-fold cross-validation. X = df.values y = labels classifiers = [ DecisionTreeClassifier(), LinearSVC(), LogisticRegression() ] for clf in classifiers: scores = cross_val_score(clf, X, y, cv=10) print("Accuracy: %0.4f (+/- %0.2f)" % (scores.mean(), scores.std() * 2))  As you can see, I am using the default values for every classifier—no grid search or other technique for finding better hyperparameters for now. Instead, I want to see how these classifiers perform out of the box. # Astonishing Results The initial test resulted in the following accuracy values: 77.84% (standard deviation 0.16) for decision trees, 64.14% (standard deviation 0.45) for support vector machines, and 55.54% (standard deviation 0.49) for logistic regression. At least the first result is highly astonishing—without any adjustments to the classifier whatsoever, we obtain a performance that is en par with more complex learning strategies! Recall that Yanardag and Vishwanathan adjusted the hyperparameters during training, while this approach merely used the default values. Let us thus focus on the decision tree classifier first and figure out why it performs the way it did. To this end, let us take a look at what the selected features look like and how important they are. To do this, I just added a single training instance, based on standard test–train split of the decision tree classifier: X_train, X_test, y_train, y_test = train_test_split(X,y, test_size=0.20) clf = DecisionTreeClassifier() clf.fit(X_train, y_train) print(clf.feature_importances_)  This results in [ 0.18874114 0.34113588 0.47012298 ], meaning that all three features are somewhat important, with density accounting for almost 50% of the purity of a node—if you are unfamiliar with decision trees, the idea is to obtain nodes that are as “pure” as possible, meaning that there should be as few differences in class labels as possible. The density attribute appears to be important for splitting up impure nodes correctly. To get a better feeling of the feature space, let us briefly visualize it using principal component analysis: clf = PCA(n_components=2) X_ = clf.fit_transform(X) for label in set(labels): idx = y[0:,] == label plt.scatter(X_[idx, 0], X_[idx, 1], label=label, alpha=0.25) plt.legend() plt.show()  This results in the following image: Every dot in the image represents an individual graph, whereas distances in the plot roughly correspond to distances between features. The large amount of overlaps between graphs with different labels thus seems to suggest that the features are insufficient to separate the individual graphs. How does the decision tree manage to separate them, nonetheless? As it turns out, by creating a very deep and wide tree. To see this, let us export the decision tree classifier: with open("/tmp/clf.txt", "w") as f: export_graphviz(clf, f)  Here is a small depiction of the resulting tree: There is also a larger variant of this tree for you to play around with. As you can see, the tree looks relatively complicated, so it is very much tailored to the given problem. That is not to say that the tree is necessarily suffering from overfitting—we explicitly used cross-validation to prevent this. What it does show is that the simple feature space with three features, while yielding very good classification results, is not detecting the true underlying structure of the problem. The rules generated by the decision tree are artificial in the sense that they do not help us understand what makes the two groups of graphs different from each other. # Coda So, where does this leave us? We saw that we were able to perform as well as state-of-the-art methods by merely picking a simple feature space, consisting of three features, and a simple decision tree classifier. I would say that any classifier with a higher complexity should yield significant improvements over this simple baseline, both in terms of average accuracy and in terms of standard deviation. Hence, the KISS principle in machine learning: start simple and only add more complex building blocks, i.e. models, if you have established a useful baseline. You can find both the code and the data in my repository on topological machine learning. To run the analysis described in this blog post, just execute $ ./simple_network_analysis.py GML/????.gml Labels.txt


in the REDDIT-BINARY subfolder of the repository.

That is all for now, until next time!

Posted at teatime on Wednesday, October 18th, 2017 Tags: