Merge pull request #1926 from flairNLP/GH-1925-TARS

GH-1925: fixed a bug, added TARS on homepage

README.md

@@ -123,7 +123,7 @@ We provide a set of quick tutorials to get you started with the library:
 * [Tutorial 6: Loading a Dataset](/resources/docs/TUTORIAL_6_CORPUS.md)
 * [Tutorial 7: Training a Model](/resources/docs/TUTORIAL_7_TRAINING_A_MODEL.md)
 * [Tutorial 8: Training your own Flair Embeddings](/resources/docs/TUTORIAL_9_TRAINING_LM_EMBEDDINGS.md)
-* [Tutorial 9: Training a Zero Shot Text Classifier](/resources/docs/TUTORIAL_10_TRAINING_ZERO_SHOT_MODEL.md)
+* [Tutorial 9: Training a Zero Shot Text Classifier (TARS)](/resources/docs/TUTORIAL_10_TRAINING_ZERO_SHOT_MODEL.md)
 
 The tutorials explain how the base NLP classes work, how you can load pre-trained models to tag your
 text, how you can embed your text with different word or document embeddings, and how you can train your own

flair/models/text_classification_model.py

@@ -699,7 +699,13 @@ def _get_nearest_labels_for(self, labels):
                     plausible_labels.append(plausible_label)
                     plausible_label_probabilities.append( \
                         self.label_nearest_map[label][plausible_label])
-            plausible_label_probabilities /= np.sum(plausible_label_probabilities) + 1e-08
+
+            # make sure the probabilities always sum up to 1
+            plausible_label_probabilities = np.array(plausible_label_probabilities, dtype='float64')
+            plausible_label_probabilities += 1e-08
+            plausible_label_probabilities /= np.sum(plausible_label_probabilities)
+
+
             if len(plausible_labels) > 0:
                 num_samples = min(self.num_negative_labels_to_sample, len(plausible_labels))
                 sampled_negative_labels = np.random.choice(plausible_labels,