update files

app.py

@@ -113,6 +113,10 @@ def xai_attributions_html(input_text: str):
     """
 
     word_attributions = cls_explainer(input_text)
+    #remove special tokens
+    word_attributions = word_attributions[1:-1]
+    # remove strings shorter than 1 chrachter
+    word_attributions = [i for i in word_attributions if len(i[0]) > 1]
     html = cls_explainer.visualize().data
     html = html.replace("#s", "")
     html = html.replace("#/s", "")
@@ -121,7 +125,7 @@ def xai_attributions_html(input_text: str):
     return word_attributions, html
 
 
-def explanation_intro(prediction_label: str): #TODO: write docstring
+def explanation_intro(prediction_label: str):
     """
     generates model explanaiton markdown from prediction label of the model.
 
@@ -131,7 +135,7 @@ def explanation_intro(prediction_label: str): #TODO: write docstring
     Returns:
       A string
     """
-    return f"""The model predicted the given sentence as **:blue[{prediction_label}]**.
+    return f"""The model predicted the given sentence as **:blue['{prediction_label}']**.
     The figure below shows the contribution of each token to this decision.
     **:green[Green]** tokens indicate a **positive contribution**, while **:red[red]** tokens indicate a **negative** contribution.
     The **bolder** the color, the greater the value."""
@@ -150,7 +154,7 @@ def explanation_viz(prediction_label: str, word_attributions):
       A string
     """
     top_attention_word = max(word_attributions, key=itemgetter(1))[0]
-    return f"""The word **_{top_attention_word}_** is the biggest driver for the decision of the model as **:blue[{prediction_label}]**."""
+    return f"""The token **_'{top_attention_word}'_** is the biggest driver for the decision of the model as **:blue['{prediction_label}']**."""
 
 
 def word_attributions_dict_creater(word_attributions):
@@ -164,9 +168,6 @@ def word_attributions_dict_creater(word_attributions):
     Returns:
       A dictionary with the keys "word", "score", and "colors".
     """
-    word_attributions = word_attributions[1:-1]
-    # remove strings shorter than 1 chrachter
-    word_attributions = [i for i in word_attributions if len(i[0]) > 1]
     word_attributions.reverse()
     words, scores = zip(*word_attributions)
     # colorize positive and negative scores
@@ -237,10 +238,11 @@ if submit:
     st.plotly_chart(label_probs_figure, config=hide_plotly_bar)
     explanation_general = explanation_intro(prediction_label)
     st.info(explanation_general)
-    word_attributions, html = xai_attributions_html(input_text)
-    st.markdown(html, unsafe_allow_html=True)
-    explanation_specific = explanation_viz(prediction_label, word_attributions)
-    st.info(explanation_specific)
-    word_attributions_dict = word_attributions_dict_creater(word_attributions)
-    attention_score_figure = attention_score_figure_creater(word_attributions_dict)
-    st.plotly_chart(attention_score_figure, config=hide_plotly_bar)
+    with st.spinner():
+      word_attributions, html = xai_attributions_html(input_text)
+      st.markdown(html, unsafe_allow_html=True)
+      explanation_specific = explanation_viz(prediction_label, word_attributions)
+      st.info(explanation_specific)
+      word_attributions_dict = word_attributions_dict_creater(word_attributions)
+      attention_score_figure = attention_score_figure_creater(word_attributions_dict)
+      st.plotly_chart(attention_score_figure, config=hide_plotly_bar)

requirements.txt

@@ -1,6 +1,7 @@
 --find-links https://download.pytorch.org/whl/torch_stable.html
+torch==1.13.1+cpu
+streamlit==1.16.0
 accelerate
 plotly
-torch==1.13.1+cpu
 transformers
 transformers-interpret