word attributes

app.py

@@ -4,6 +4,8 @@ from huggingface_hub import hf_hub_download
 from langdetect import detect, DetectorFactory, detect_langs
 import fasttext
 from transformers import pipeline
+from transformers_interpret import ZeroShotClassificationExplainer
+import string, nltk
 
 models = {'en': 'Narsil/deberta-large-mnli-zero-cls', #'microsoft/deberta-xlarge-mnli' # English
           'de': 'Sahajtomar/German_Zeroshot', # German
@@ -38,6 +40,10 @@ classifiers = {'en': pipeline("zero-shot-classification", hypothesis_template=hy
 
 fasttext_model = fasttext.load_model(hf_hub_download("julien-c/fasttext-language-id", "lid.176.bin"))
 
+_ = nltk.download('stopwords', quiet=True)
+_ = nltk.download('wordnet', quiet=True)
+_ = nltk.download('punkt', quiet=True)
+
 def prep_examples():
     example_text1 = "Coronavirus disease (COVID-19) is an infectious disease caused by the SARS-CoV-2 virus. Most \
     people who fall sick with COVID-19 will experience mild to moderate symptoms and recover without special treatment. \
@@ -156,6 +162,23 @@ def sequence_to_classify(sequence, labels):
         str(datetime.datetime.now()),
         sequence,
         predicted_labels))
+    
+    # Explain word attributes
+    stop_words =  nltk.corpus.stopwords.words('english')
+    puncts = list(string.punctuation)
+        
+    model_expl = ZeroShotClassificationExplainer(classifier.model, classifier.tokenizer)
+    response_expl = model_expl(sequence, labels, hypothesis_template="This example is {}.")
+    
+    if len(labels_pred) == 1:
+        response_expl = response_expl[model_expl.predicted_label]
+
+    for key in response_expl:
+        for idx, elem in enumerate(response_expl[key]):
+            if elem[0] in stop_words:
+                del response_expl[key][idx]
+    
+    print(response_expl)
 
     return clean_output
 

requirements.txt

@@ -3,4 +3,6 @@ sentence-transformers
 torch
 langdetect
 fasttext
+transformers_interpret
+nltk