metrics for binary classification scoring

evaluation/evaluate.py

@@ -13,6 +13,9 @@ def evaluation_metrics(y, y_):
             macroK(y, y_),
             microK(y, y_),
             # macroAcc(y, y_),
+            microAcc(
+                y, y_
+            ),  # TODO: we're using micro-averaging for accuracy, it is == to accuracy_score on binary classification
         )
 
 
@@ -32,10 +35,12 @@ def log_eval(l_eval, phase="training", verbose=True):
         print(f"\n[Results {phase}]")
     metrics = []
     for lang in l_eval.keys():
-        macrof1, microf1, macrok, microk = l_eval[lang]
-        metrics.append([macrof1, microf1, macrok, microk])
+        macrof1, microf1, macrok, microk, microAcc = l_eval[lang]
+        metrics.append([macrof1, microf1, macrok, microk, microAcc])
         if phase != "validation":
-            print(f"Lang {lang}: macro-F1 = {macrof1:.3f} micro-F1 = {microf1:.3f}")
+            print(
+                f"Lang {lang}: macro-F1 = {macrof1:.3f} micro-F1 = {microf1:.3f} acc = {microAcc:.3f}"
+            )
     averages = np.mean(np.array(metrics), axis=0)
     if verbose:
         print(

evaluation/metrics.py

@@ -239,3 +239,7 @@ def microK(true_labels, predicted_labels):
 
 def macroAcc(true_labels, predicted_labels):
     return macro_average(true_labels, predicted_labels, accuracy)
+
+
+def microAcc(true_labels, predicted_labels):
+    return micro_average(true_labels, predicted_labels, accuracy)