Fix lint errors, fix boolean mask used in test_predictor, which gives different results on different versions of pandas

Author: timodonnell

pepnet/synthetic_data/subsequence.py

@@ -42,8 +42,8 @@ def synthetic_peptides_by_subsequence(
     if not isinstance(lengths, dict):
         lengths = dict((length, 1.0) for length in lengths)
 
-    lengths = pandas.Series(lengths)
-    lengths /= len(lengths)
+    lengths_series = pandas.Series(lengths)
+    lengths_series /= len(lengths)
 
     num_binders = int(round(num_peptides * fraction_binders))
     num_non_binders = num_peptides - num_binders
@@ -52,15 +52,20 @@ def synthetic_peptides_by_subsequence(
     peptides = []
 
     # Generate non-binders
-    for (length, weight) in lengths.items():
+    for (length, weight) in lengths_series.items():
         peptides.extend(
             random_peptides(round(weight * num_non_binders), round(length)))
 
     for binding_core in binding_subsequences:
         # Generate binders
-        lengths_binders = lengths.ix[lengths.index >= len(binding_core)]
+        lengths_binders = lengths_series.ix[
+            lengths_series.index >= len(binding_core)
+        ]
         normalized_lengths_binders = (
-            lengths_binders / lengths_binders.sum() / len(binding_subsequences))
+            lengths_binders /
+            lengths_binders.sum() /
+            len(binding_subsequences))
+
         for (length, weight) in normalized_lengths_binders.items():
             if length >= len(binding_core):
                 num_peptides_to_make = int(round(weight * num_binders))

pylintrc

@@ -1,4 +1,4 @@
 [TYPECHECK]
 # Without ignoring this, we get errors like:
 # E:249,20: Module 'numpy' has no 'nan' member (no-member)
-ignored-modules = numpy
+ignored-modules = numpy,numpy.random

test/test_predictor.py

@@ -85,8 +85,8 @@ def test_predictor_on_more_data():
     predictor.fit(
         {"x": train_df.index.values}, train_df.binder.values, epochs=20)
     y_pred = predictor.predict({"x": test_df.index.values})['y']
-    binder_mean_pred = y_pred[test_df.binder].mean()
-    nonbinder_mean_pred = y_pred[~test_df.binder].mean()
+    binder_mean_pred = y_pred[test_df.binder > 0].mean()
+    nonbinder_mean_pred = y_pred[test_df.binder == 0].mean()
     print(binder_mean_pred, nonbinder_mean_pred)
     assert binder_mean_pred > nonbinder_mean_pred * 2, (
         binder_mean_pred, nonbinder_mean_pred)