update

util/evaluation.py

@@ -7,6 +7,7 @@ from scipy.stats import spearmanr, pearsonr, kendalltau, entropy
 from scipy.spatial.distance import jensenshannon
 from scipy.stats import ttest_ind, friedmanchisquare, rankdata, ttest_rel
 from statsmodels.stats.multicomp import pairwise_tukeyhsd
+from scipy.stats import ttest_1samp
 
 # def bootstrap_t_test(data1, data2, num_bootstrap=1000):
 #     """Perform a bootstrapped t-test."""
@@ -24,18 +25,26 @@ from statsmodels.stats.multicomp import pairwise_tukeyhsd
 #     p_value = np.sum(np.abs(t_stats) >= np.abs(observed_t_stat)) / num_bootstrap
 #     return observed_t_stat, p_value
 
+
 def bootstrap_t_test(data1, data2, num_bootstrap=1000):
-    """Perform a bootstrapped paired t-test."""
-    observed_t_stat, _ = ttest_rel(data1, data2)
+    """Perform a bootstrapped paired t-test for mean difference being zero."""
+    # Calculate the observed differences between paired samples
     differences = data1 - data2
+    # Compute the observed t-statistic for the differences
+    observed_t_stat, _ = ttest_1samp(differences, 0)
+
     t_stats = []
 
     for _ in range(num_bootstrap):
-        # Resample with replacement
+        # Resample the differences with replacement
         resampled_diffs = np.random.choice(differences, size=len(differences), replace=True)
-        t_stat, _ = ttest_rel(resampled_diffs, np.zeros(len(resampled_diffs)))
+        # Perform a one-sample t-test on the resampled differences against zero
+        t_stat, _ = ttest_1samp(resampled_diffs, 0)
+        # Append the t-statistic to the list
         t_stats.append(t_stat)
 
+    # Calculate the p-value as the proportion of bootstrap t-statistics
+    # that are as extreme as or more extreme than the observed t-statistic
     p_value = np.sum(np.abs(t_stats) >= np.abs(observed_t_stat)) / num_bootstrap
     return observed_t_stat, p_value