Update app.py

app.py

@@ -1,6 +1,7 @@
 # Import packages:
 
 import numpy as np
+import pandas as pd
 import matplotlib.pyplot as plt
 import re
 # tensorflow imports:
@@ -10,16 +11,7 @@ from tensorflow.keras import losses
 from tensorflow.keras import layers
 from tensorflow.keras.layers.experimental import preprocessing
 from tensorflow.keras.optimizers import RMSprop
-# # keras imports:
-from keras.models import Model
-from keras.layers import LSTM, Activation, Dense, Dropout, Input, Embedding, RepeatVector, TimeDistributed
-from keras.preprocessing.text import Tokenizer
-from keras_preprocessing import sequence
-from tensorflow.keras.utils import to_categorical
-from keras.callbacks import EarlyStopping
-from keras.models import Sequential
-from keras import layers
-from keras.backend import clear_session
+
 import pickle
 import gradio as gr
 import yake
@@ -28,6 +20,15 @@ from spacy import displacy
 import streamlit as st
 import spacy_streamlit
 nlp = spacy.load('en_core_web_sm')
+import torch
+import tensorflow as tf
+from transformers import RobertaTokenizer, RobertaModel, AutoModelForSequenceClassification, TFAutoModelForSequenceClassification
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
+tokenizer = AutoTokenizer.from_pretrained("paragon-analytics/roberta_persuade")
+model = AutoModelForSequenceClassification.from_pretrained("paragon-analytics/roberta_persuade")
+
+# para_tokenizer = AutoTokenizer.from_pretrained("paragon-analytics/t5_para")
+# para_model = AutoModelForSeq2SeqLM.from_pretrained("paragon-analytics/t5_para")
 
 kw_extractor = yake.KeywordExtractor()
 custom_kw_extractor = yake.KeywordExtractor(lan="en", n=2, dedupLim=0.2, top=10, features=None)
@@ -35,22 +36,32 @@ custom_kw_extractor = yake.KeywordExtractor(lan="en", n=2, dedupLim=0.2, top=10,
 max_words = 2000
 max_len = 111
 
+from transformers_interpret import SequenceClassificationExplainer
+cls_explainer = SequenceClassificationExplainer(
+    model,
+    tokenizer)
+    
 # load the model from disk
-filename = 'lstm_model.sav'
-lmodel = pickle.load(open(filename, 'rb'))
+#filename = 'resil_lstm_model.sav'
+#lmodel = pickle.load(open(filename, 'rb'))
 
 # load the model from disk
-filename = 'tokenizer.pickle'
-tok = pickle.load(open(filename, 'rb'))
+#filename = 'tokenizer.pickle'
+#tok = pickle.load(open(filename, 'rb'))
 
 def process_final_text(text):
     X_test = str(text).lower()
-    l = []
-    l.append(X_test)
-    test_sequences = tok.texts_to_sequences(l)
-    test_sequences_matrix = sequence.pad_sequences(test_sequences,maxlen=max_len)
-    lstm_prob = lmodel.predict(test_sequences_matrix.tolist()).flatten()
-    lstm_pred = np.where(lstm_prob>=0.5,1,0)
+    #l = []
+    #l.append(X_test)
+    #test_sequences = tok.texts_to_sequences(l)
+    #test_sequences_matrix = sequence.pad_sequences(test_sequences,maxlen=max_len)
+    #lstm_prob = lmodel.predict(test_sequences_matrix.tolist()).flatten()
+    #lstm_pred = np.where(lstm_prob>=0.5,1,0)
+    
+    encoded_input = tokenizer(X_test, return_tensors='pt')
+    output = model(**encoded_input)
+    scores = output[0][0].detach().numpy()
+    scores = tf.nn.softmax(scores)
     
     # Get Keywords:
     keywords = custom_kw_extractor.extract_keywords(X_test)
@@ -80,36 +91,50 @@ def process_final_text(text):
         + sp_html
         + ""
     )
-    return {"Persuasive": float(lstm_prob[0]), "Non-Persuasive": 1-float(lstm_prob[0])},keywords,NER
     
-def main(prob1,prob2,sol1,sol2,inv,act):
-    text = str(prob1) + " " + str(prob2) + " " + str(sol1) + " " + str(sol2) + " " + str(inv) + " " + str(act)
+    # Transformer Interpret:
+    word_attributions = cls_explainer(X_test)
+    letter = []
+    score = []
+    for i in word_attributions:
+        if i[1]>0.5:
+            a = "++"
+        elif (i[1]<=0.5) and (i[1]>0.1):
+            a = "+"
+        elif (i[1]>=-0.5) and (i[1]<-0.1):
+            a = "-"
+        elif i[1]<-0.5:
+            a = "--"
+        else: 
+            a = "NA"
+    
+        letter.append(i[0])
+        score.append(a)
+    
+    word_attributions = [(letter[i], score[i]) for i in range(0, len(letter))]
+
+    # # Paraphraser:
+    # batch = para_tokenizer(X_test, return_tensors='pt')
+    # generated_ids = para_model.generate(batch['input_ids'])
+    # para_list = para_tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
+
+    return {"Persuasive": float(scores.numpy()[1]), "Non-Persuasive": float(scores.numpy()[0])},keywords,NER,word_attributions
+    
+def main(prob1):
+    text = str(prob1)
     obj = process_final_text(text)
-    return obj[0],obj[1],obj[2]
+    return obj[0],obj[1],obj[2],obj[3]
     
-title = "Welcome to **PersuAID** 🪐"
+title = "Welcome to **Persuade** 🪐"
 description1 = """
-It is difficult to write persuasive product descriptions. It could take time and money to create a good one. PersuAID uses a template to organize your thoughts when writing a persuasive product description. It's AI model is trained on tens of thousands of product descriptions. Why don't you give it a try? Just add your text and hit Create & Analyze ✨   
-"""
-
-description2 = """
-Although encouraged, you don't have to fill all the boxes. Just try the ones that matter to you. After getting your first score, modify your answers and hit Create & Analyze again 🤞
-""" 
+This app takes text (up to a few sentences) and predicts to what extent the text contains persuasive content.""" 
 
 with gr.Blocks(title=title) as demo:
     gr.Markdown(f"## {title}")
-    gr.Markdown("""![marketing](file/marketing.jpg)""")
     gr.Markdown(description1)
     gr.Markdown("""---""")
-    gr.Markdown(description2)
-    gr.Markdown("""---""")
-    prob1 = gr.Textbox(label="A great product description is the one that solves an important problem. What is the single, most important problem that your product solves?",lines=2, placeholder="Type it here ...")
-    prob2 = gr.Textbox(label="Are there any other problems your product solves? If none, leave blank.",lines=2, placeholder="Type them here, ...")
-    sol1 = gr.Textbox(label="What solution does your product offer to the main problem you described first?",lines=2, placeholder="Type your answer here ...")
-    sol2 = gr.Textbox(label="How does your product solves the problem?",lines=2, placeholder="Type your answer here ...")
-    inv = gr.Textbox(label="Now it's time to invite your audience to take action. Encourage them to try your product 💖",lines=2, placeholder="Type your invitation here ...")
-    act = gr.Textbox(label="And finally, describe (short sentence) what action you want them to take. This could also be your product description as a punchline 💪",lines=2, placeholder="Type the desired action/ outcome/ punchline here ...")
-    submit_btn = gr.Button("Create & Analyze")
+    prob1 = gr.Textbox(label="Enter Your Text Here:",lines=2, placeholder="Type it here ...")
+    submit_btn = gr.Button("Analyze")
     #text = gr.Textbox(label="Text:",lines=2, placeholder="Please enter text here ...")
     #submit_btn2 = gr.Button("Analyze")
 
@@ -119,17 +144,18 @@ with gr.Blocks(title=title) as demo:
         color_map={"+++": "royalblue","++": "cornflowerblue",
          "+": "lightsteelblue", "NA":"white"})
         NER = gr.HTML(label = 'NER:')
+        intp =gr.HighlightedText(label="Word Scores",
+        combine_adjacent=False).style(color_map={"++": "darkgreen","+": "green", 
+                                                "--": "darkred",
+                                                "-": "red", "NA":"white"})
 
     submit_btn.click(
         main,
-        [prob1,prob2,sol1,sol2,inv,act],
-        [label,impplot,NER], api_name="PrsTalk"
+        [prob1],
+        [label,impplot,NER,intp], api_name="ResText"
     )
 
-    
-    
-    gr.Markdown("### Click on any of the examples below to see how it works:")
-    gr.Examples([["It is difficult to write persuasive product descriptions.", "It could take time and money to create a good one.", "PersuAID uses a template to organize your thoughts when writing a persuasive product description.", "It's AI model is trained on tens of thousands of product descriptions.", "Why don't you give it a try?", "Just add your text and hit Create & Analyze ✨"], ["What is performance?", "Zero to Sixty or Sixty to Zero? How a car performs a quarter mile or a quarter century? Is performance about the joy of driving or the importance of surviving?",
- "To us performance is not about doing one thing well ...", "it is about doing everything well .. because in the end everything matters.", "Performance without compromise.", "That is what drives you..... Mercedes Benz"], ["Talking to your friends about their problems with drugs and alcohol might not be easy.", "",  "", "", "Courage.", "The anti-drug."], ["When experiencing depression, I couldn't get out of bed or focus.", "My goals and dreams drifted away.", "I got my life back through the help of a support group of people who had been where I was.", "They helped me see that my life wasn't limited that I could still achieve my dreams.","We've been there.", "We can help. Call 800-826-3632 to learn more."], ["Up to 6 million homeless animals enter shelters nationwide every year.", "Some abandoned, some sick or injured they're just looking for a place to call home.", "When you rescue a shelter animal you won't just be making a difference in their life, you could be making a difference in your own.", "", "Visit your local animal shelter today and save a life.", "HumaneSociety.org"], ["You always dreamed of living in a place like this,", "But the land can be challenging.", "That's where our expertise comes in.", "Your John Deere dealer will find the best solution to get the most for your money and the most out of your property...", "Because we know you love your land...every last bit of it.", "That's how we run... and nothin' runs like a Deere."]], [prob1,prob2,sol1,sol2,inv,act], [label,impplot,NER], main, cache_examples=True)
+    gr.Markdown("### Click on any of the examples below to see to what extent they contain resilience messaging:")
+    gr.Examples([["Please stay at home and avoid unnecessary trips."],["Please stay at home and avoid unnecessary trips. We will survive this."],["We will survive this."],["Watch today’s news briefing with the latest updates on COVID-19 in Connecticut."],["So let's keep doing what we know works. Let's stay strong, and let's beat this virus. I know we can, and I know we can come out stronger on the other side."],["It is really wonderful how much resilience there is in human nature. Let any obstructing cause, no matter what, be removed in any way, even by death, and we fly back to first principles of hope and enjoyment."],["Resilience is accepting your new reality, even if it’s less good than the one you had before. You can fight it, you can do nothing but scream about what you’ve lost, or you can accept that and try to put together something that’s good."],["You survived all of the days you thought you couldn't, never underestimate your resilience."],["Like tiny seeds with potent power to push through tough ground and become mighty trees, we hold innate reserves of unimaginable strength. We are resilient."]], [prob1], [label,impplot,NER,intp], main, cache_examples=True)
     
 demo.launch()