Return Python float instead of numpy.float64 in sklearn metrics (#2612)

Author: lewtun

metrics/accuracy/accuracy.py

@@ -83,5 +83,7 @@ def _info(self):
 
     def _compute(self, predictions, references, normalize=True, sample_weight=None):
         return {
-            "accuracy": accuracy_score(references, predictions, normalize=normalize, sample_weight=sample_weight),
+            "accuracy": accuracy_score(
+                references, predictions, normalize=normalize, sample_weight=sample_weight
+            ).tolist(),
         }

metrics/f1/f1.py

@@ -106,5 +106,5 @@ def _compute(self, predictions, references, labels=None, pos_label=1, average="b
                 pos_label=pos_label,
                 average=average,
                 sample_weight=sample_weight,
-            ),
+            ).tolist(),
         }

metrics/glue/glue.py

@@ -81,21 +81,21 @@
 
 
 def simple_accuracy(preds, labels):
-    return (preds == labels).mean()
+    return (preds == labels).mean().tolist()
 
 
 def acc_and_f1(preds, labels):
     acc = simple_accuracy(preds, labels)
-    f1 = f1_score(y_true=labels, y_pred=preds)
+    f1 = f1_score(y_true=labels, y_pred=preds).tolist()
     return {
         "accuracy": acc,
         "f1": f1,
     }
 
 
 def pearson_and_spearman(preds, labels):
-    pearson_corr = pearsonr(preds, labels)[0]
-    spearman_corr = spearmanr(preds, labels)[0]
+    pearson_corr = pearsonr(preds, labels)[0].tolist()
+    spearman_corr = spearmanr(preds, labels)[0].tolist()
     return {
         "pearson": pearson_corr,
         "spearmanr": spearman_corr,

metrics/indic_glue/indic_glue.py

@@ -74,12 +74,12 @@
 
 
 def simple_accuracy(preds, labels):
-    return (preds == labels).mean()
+    return (preds == labels).mean().tolist()
 
 
 def acc_and_f1(preds, labels):
     acc = simple_accuracy(preds, labels)
-    f1 = f1_score(y_true=labels, y_pred=preds)
+    f1 = f1_score(y_true=labels, y_pred=preds).tolist()
     return {
         "accuracy": acc,
         "f1": f1,
@@ -99,7 +99,7 @@ def precision_at_10(en_sentvecs, in_sentvecs):
     actual = np.array(range(n))
     preds = sim.argsort(axis=1)[:, :10]
     matches = np.any(preds == actual[:, None], axis=1)
-    return matches.mean()
+    return matches.mean().tolist()
 
 
 @datasets.utils.file_utils.add_start_docstrings(_DESCRIPTION, _KWARGS_DESCRIPTION)

metrics/matthews_correlation/matthews_correlation.py

@@ -82,5 +82,5 @@ def _info(self):
 
     def _compute(self, predictions, references, sample_weight=None):
         return {
-            "matthews_correlation": matthews_corrcoef(references, predictions, sample_weight=sample_weight),
+            "matthews_correlation": matthews_corrcoef(references, predictions, sample_weight=sample_weight).tolist(),
         }

metrics/precision/precision.py

@@ -108,5 +108,5 @@ def _compute(self, predictions, references, labels=None, pos_label=1, average="b
                 pos_label=pos_label,
                 average=average,
                 sample_weight=sample_weight,
-            ),
+            ).tolist(),
         }

metrics/recall/recall.py

@@ -108,5 +108,5 @@ def _compute(self, predictions, references, labels=None, pos_label=1, average="b
                 pos_label=pos_label,
                 average=average,
                 sample_weight=sample_weight,
-            ),
+            ).tolist(),
         }

metrics/super_glue/super_glue.py

@@ -107,12 +107,12 @@
 
 
 def simple_accuracy(preds, labels):
-    return (preds == labels).mean()
+    return (preds == labels).mean().tolist()
 
 
 def acc_and_f1(preds, labels, f1_avg="binary"):
     acc = simple_accuracy(preds, labels)
-    f1 = f1_score(y_true=labels, y_pred=preds, average=f1_avg)
+    f1 = f1_score(y_true=labels, y_pred=preds, average=f1_avg).tolist()
     return {
         "accuracy": acc,
         "f1": f1,
@@ -138,9 +138,9 @@ def evaluate_multirc(ids_preds, labels):
         f1s.append(f1)
         em = int(sum([p == l for p, l in preds_labels]) == len(preds_labels))
         ems.append(em)
-    f1_m = sum(f1s) / len(f1s)
+    f1_m = (sum(f1s) / len(f1s)).tolist()
     em = sum(ems) / len(ems)
-    f1_a = f1_score(y_true=labels, y_pred=[id_pred["prediction"] for id_pred in ids_preds])
+    f1_a = f1_score(y_true=labels, y_pred=[id_pred["prediction"] for id_pred in ids_preds]).tolist()
     return {"exact_match": em, "f1_m": f1_m, "f1_a": f1_a}