Update app.py

app.py

@@ -28,29 +28,35 @@ cats = ["Computer Science", "Economics", "Electrical Engineering",
 
 def predict(outputs):
     top = 0
-    probs = nn.functional.softmax(outputs, dim=1).tolist()[0]
+    temp = 0.5
+    probs = nn.functional.softmax(outputs / temp, dim=1).tolist()[0]
     
     top_cats = []
     top_probs = []
 
     first = True
+    write_cs = False
     for prob, cat in sorted(zip(probs, cats), reverse=True):
         if first:
             if cat == "Computer Science":
-                st.write("Today everything is connected with Computer Science")
+                write_cs = True
             first = False
         if top < 95:
             percent = prob * 100
             top += percent
             top_cats.append(cat)
-            top_probs.append(round(percent, 1))
-    return pd.DataFrame(top_probs, index=top_cats, columns=['Percent'])
+            top_probs.append(str(round(percent, 1)))
+    res = pd.DataFrame(top_probs, index=top_cats, columns=['Percent'])
+    st.write(res)
+
+    if write_cs:
+        st.write("Today everything is connected with Computer Science")
 
 tokenizer = RobertaTokenizer.from_pretrained("roberta-large-mnli")
 model = init_model()
 
 st.title("Article classifier")
-st.markdown("<img width=200px src='https://lionbridge.ai/wp-content/uploads/2020/09/2020-09-08_text-classification-tools-services.jpg'>", unsafe_allow_html=True)
+st.markdown("<img width=500px src='https://lionbridge.ai/wp-content/uploads/2020/09/2020-09-08_text-classification-tools-services.jpg'>", unsafe_allow_html=True)
 st.markdown("### Title")
 
 title = st.text_area("*Enter title (required)", height=20)
@@ -67,5 +73,4 @@ else:
                           max_length=1024, truncation=True)
     with torch.no_grad():
         outputs = model(**encoded_input).pooler_output[:, 0, :]
-        res = predict(outputs)
-        st.write(res)
+        predict(outputs)