better handling of bootstrap plotting when infinities are present (#72)

Author: josesho

dabest/_classes.py

@@ -460,7 +460,7 @@ def __init__(self, control, test, effect_size,
          'statistic_wilcoxon': nan}
         """
 
-        from numpy import array, isnan
+        from numpy import array, isnan, isinf
         from numpy import sort as npsort
         from numpy.random import choice, seed
 
@@ -522,6 +522,20 @@ def __init__(self, control, test, effect_size,
                             control, test, is_paired, effect_size,
                             resamples, random_seed)
         self.__bootstraps = npsort(bootstraps)
+        
+        # Added in v0.2.6.
+        # Raises a UserWarning if there are any infiinities in the bootstraps.
+        num_infinities = len(self.__bootstraps[isinf(self.__bootstraps)])
+        
+        if num_infinities > 0:
+            warn_msg = "There are {} bootstrap(s) that are not defined. "\
+            "This is likely due to smaple sample sizes. "\
+            "The values in a bootstrap for a group will be more likely "\
+            "to be all equal, with a resulting variance of zero. "\
+            "The computation of Cohen's d and Hedges' g thus "\
+            "involved a division by zero. "
+            warnings.warn(warn_msg.format(num_infinities), 
+                          category=UserWarning)
 
         self.__bias_correction = ci2g.compute_meandiff_bias_correction(
                                     self.__bootstraps, self.__difference)

dabest/_stats_tools/confint_2group_diff.py

@@ -159,7 +159,24 @@ def compute_bootstrapped_diff(x0, x1, is_paired, effect_size,
 
     # reset seed
     np.random.seed()
-
+    
+    # check whether there are any infinities in the bootstrap,
+    # which likely indicates the sample sizes are too small as
+    # the computation of Cohen's d and Hedges' g necessitated 
+    # a division by zero.
+    # Added in v0.2.6.
+    
+    # num_infinities = len(out[np.isinf(out)])
+    # print(num_infinities)
+    # if num_infinities > 0:
+    #     warn_msg = "There are {} bootstraps that are not defined. "\
+    #     "This is likely due to smaple sample sizes. "\
+    #     "The values in a bootstrap for a group will be more likely "\
+    #     "to be all equal, with a resulting variance of zero. "\
+    #     "The computation of Cohen's d and Hedges' g will therefore "\
+    #     "involved a division by zero. "
+    #     warnings.warn(warn_msg.format(num_infinities), category="UserWarning")
+        
     return out
 
 

dabest/_stats_tools/effsize.py

@@ -217,11 +217,13 @@ def cohens_d(control, test, is_paired=False):
         # assume the two arrays are ordered already.
         delta = test - control
         M = np.mean(delta)
-        return M / average_sd
+        divisor = average_sd
 
     else:
         M = np.mean(test) - np.mean(control)
-        return M / pooled_sd
+        divisor = pooled_sd
+        
+    return M / divisor
 
 
 

dabest/plotter.py

@@ -471,7 +471,8 @@ def EffectSizeDataFramePlotter(EffectSizeDataFrame, **plot_kwargs):
         current_ci_high   = results.bca_high[j]
 
         # Create the violinplot.
-        v = contrast_axes.violinplot(current_bootstrap,
+        # New in v0.2.6: drop negative infinities before plotting.
+        v = contrast_axes.violinplot(current_bootstrap[~np.isinf(current_bootstrap)],
                                      positions=[tick],
                                      **violinplot_kwargs)
         # Turn the violinplot into half, and color it the same as the swarmplot.