1. Guest lecture: Feature Selection Alan Qi yuanqi@mit.edu Dec 2, 2004

2. Outline Problems Overview of feature selection (FS) –Filtering: correlation & information criteria –Wrapper approach: greedy FS & regularization Classical Bayesian feature selection New Bayesian approach: predictive-ARD

3. Feature selection Gene expression: thousands of gene measurements Documents: “bag of words” model with more than 10,000 words Images: histograms, colors, wavelet coefficients, etc. Task: find a small subset of features for prediction

4. Gene Expression Classification Task: Classify gene expression datasets into different categories, e.g., normal v.s. cancer Challenge: Thousands of genes measured in the micro-array data. Only a small subset of genes are probably correlated with the classification task.

5. Filtering approach Feature ranking based on sensible criteria: –Correlation between features and labels –Mutual information between features and labels

6. Wrapper Approach Assess subsets of variables according to their usefulness to a given predictor. A combinatorial problem: 2 K combinations given K features. -Sequentially adding/removing features: Sequential Forward Selection (SFS), Backward Sequential Selection (SBS). -Recursively adding/removing features: Sequential Forward Floating Selection (SFFS) (When to stop? Overfitting?) -Regularization: use sparse prior to enhance the sparsity of a trained predictor (classifier).

7. Regularization Regularization: combining the fit to the data and a penalty for complexity. Minimizing the following Labels: t = [t 1, t 1, …, t N ] Inputs: X = [x 1, x 1, …, x N ] Parameters: w Likelihood for the data set (For classification):

8. Bayesian feature selection Background –Bayesian classification model –Automatic relevance determination (ARD) Risk of Overfitting by optimizing hyperparameters Predictive ARD by expectation propagation (EP): –Approximate prediction error –EP approximation Experiments Conclusions

9. Motivation Task 1: Classify high dimensional datasets with many irrelevant features, e.g., normal v.s. cancer microarray data. Task 2: Sparse Bayesian kernel classifiers for fast test performance.

10. Bayesian Classification Model Prior of the classifier w: Labels: t inputs: X parameters: w Likelihood for the data set: Where is a cumulative distribution function for a standard Gaussian.

11. Evidence and Predictive Distribution The evidence, i.e., the marginal likelihood of the hyperparameters : The predictive posterior distribution of the label for a new input :

12. Automatic Relevance Determination (ARD) Give the classifier weight independent Gaussian priors whose variance,, controls how far away from zero each weight is allowed to go: Maximize, the marginal likelihood of the model, with respect to. Outcome: many elements of go to infinity, which naturally prunes irrelevant features in the data.

13. Two Types of Overfitting Classical Maximum likelihood: –Optimizing the classifier weights w can directly fit noise in the data, resulting in a complicated model. Type II Maximum likelihood (ARD): –Optimizing the hyperparameters corresponds to choosing which variables are irrelevant. Choosing one out of exponentially many models can also overfit if we maximize the model marginal likelihood.

14. Risk of Optimizing X: Class 1 vs O: Class 2 Evd-ARD-1 Evd-ARD-2 Bayes Point

15. Outline Background –Bayesian classification model –Automatic relevance determination (ARD) Risk of Overfitting by optimizing hyperparameters Predictive ARD by expectation propagation (EP): –Approximate prediction error –EP approximation Experiments Conclusions

16. Predictive-ARD Choosing the model with the best estimated predictive performance instead of the most probable model. Expectation propagation (EP) estimates the leave-one-out predictive performance without performing any expensive cross-validation.

17. Estimate Predictive Performance Predictive posterior given a test data point EP can estimate predictive leave-one-out error probability where q ( w| t \ i ) is the approximate posterior of leaving out the i th label. EP can also estimate predictive leave-one-out error count

18. Expectation Propagation in a Nutshell Approximate a probability distribution by simpler parametric terms: Each approximation term lives in an exponential family (e.g. Gaussian)

19. EP in a Nutshell Three key steps: Deletion Step: approximate the “leave-one-out” predictive posterior for the i th point: Minimizing the following KL divergence by moment matching: Inclusion: The key observation: we can use the approximate predictive posterior, obtained in the deletion step, for model selection. No extra computation!

20. Comparison of different model selection criteria for ARD training 1 st row: Test error 2 nd row: Estimated leave-one-out error probability 3 rd row: Estimated leave-one-out error counts 4 th row: Evidence (Model marginal likelihood) 5 th row: Fraction of selected features The estimated leave-one-out error probabilities and counts are better correlated with the test error than evidence and sparsity level.

21. Gene Expression Classification Task: Classify gene expression datasets into different categories, e.g., normal v.s. cancer Challenge: Thousands of genes measured in the micro-array data. Only a small subset of genes are probably correlated with the classification task.

22. Classifying Leukemia Data The task: distinguish acute myeloid leukemia (AML) from acute lymphoblastic leukemia (ALL). The dataset: 47 and 25 samples of type ALL and AML respectively with 7129 features per sample. The dataset was randomly split 100 times into 36 training and 36 testing samples.

23. Classifying Colon Cancer Data The task: distinguish normal and cancer samples The dataset: 22 normal and 40 cancer samples with 2000 features per sample. The dataset was randomly split 100 times into 50 training and 12 testing samples. SVM results from Li et al. 2002

24. Bayesian Sparse Kernel Classifiers Using feature/kernel expansions defined on training data points: Predictive-ARD-EP trains a classifier that depends on a small subset of the training set. Fast test performance.

25. Test error rates and numbers of relevance or support vectors on breast cancer dataset. 50 partitionings of the data were used. All these methods use the same Gaussian kernel with kernel width = 5. The trade-off parameter C in SVM is chosen via 10- fold cross-validation for each partition.

26. Test error rates on diabetes data 100 partitionings of the data were used. Evidence and Predictive ARD-EPs use the Gaussian kernel with kernel width = 5.

27. Summary Two kinds of feature selection methods: –Filtering and wrapper methods Classical Bayesian feature selection: –Excellent classical approach: Tuning prior to prune features. –However, maximizing marginal likelihood can lead to overfitting in the model space if there are a lot of features. New Bayesian approach: Predictive-ARD, which focus on the prediction performance. –feature selection –sparse kernel learning