fixed tests

Author: Tomasz Latkowski

experiments/experiment.py

@@ -1,8 +1,7 @@
 import tensorflow as tf
 
-from experiments.classifier import NeuralNetworkClassifier
-from methods.selection_wrapper import SelectionWrapper
 from methods.selection import fisher, feature_correlation_with_class, t_test, random
+from methods.selection_wrapper import SelectionWrapper
 
 methods = {
     'fisher': fisher,
@@ -15,7 +14,6 @@
 class Experiment:
 
     def __init__(self, experiment_config, num_instances, classifier, dataset):
-
         selection_method = methods[experiment_config['SELECTION']['method']]
         num_features = int(experiment_config['SELECTION']['num_features'])
         hidden_sizes = int(experiment_config['CLASSIFIER']['hidden_sizes'])

run.py

@@ -70,4 +70,4 @@ def main():
 
 
 if __name__ == '__main__':
-    main()
+    main()

tests/test_fisher.py

@@ -1,7 +1,8 @@
 import numpy as np
 import tensorflow as tf
 
-from methods.selection import fisher, selection_wrapper
+from methods.selection import fisher
+from methods.selection_wrapper import SelectionWrapper
 
 
 class TestFisherSelection(tf.test.TestCase):
@@ -28,10 +29,11 @@ def testFisherPickFirstSignificantFeature(self):
 
             num_instances = [2, 2]
             top_k = 1
-            _, actual_most_significant_features = test_session.run(selection_wrapper(data,
-                                                                                     num_instances,
-                                                                                     fisher,
-                                                                                     num_features=top_k))
+            selection_wrapper = SelectionWrapper(data,
+                                                 num_instances,
+                                                 fisher,
+                                                 num_features=top_k)
+            actual_most_significant_features = test_session.run(selection_wrapper.selected_data)
             correct_most_significant_features = [[2.], [4.], [6.], [6.]]
 
             self.assertAllEqual(actual_most_significant_features, correct_most_significant_features)

utils/statistics.py

@@ -12,5 +12,33 @@ def pearson_correlation(x1, x2):
     return l / p
 
 
-def f_test():
-    pass
+def f_test(data, num_instances):
+    """
+    Performs F-statistic between the genes and the classification variable h
+    as the score of maximum relevance.
+
+    :param data:
+    :param num_instances:
+    :return:
+    """
+
+    data = tf.convert_to_tensor(data)
+    class1, class2 = tf.split(data, num_instances)
+    K = 2
+    with tf.name_scope('f_statistic'):
+        mean1, var1 = tf.nn.moments(class1, axes=0)
+        mean2, var2 = tf.nn.moments(class2, axes=0)
+        mean, var = tf.nn.moments(data, axes=0)
+
+        pooled_var = pooled_variance(data, num_instances)
+        tf.reduce_sum(((mean1 - mean) + (mean2 - mean))/(K-1))/pooled_var
+
+
+def pooled_variance(data, num_instances):
+    K = len(num_instances)
+    n = sum(num_instances)
+
+    class1, class2 = tf.split(data, num_instances)
+    mean1, var1 = tf.nn.moments(class1, axes=0)
+    mean2, var2 = tf.nn.moments(class2, axes=0)
+    return -1