app.py

import streamlit as st
import gradio as gr
import shap
import numpy as np
import scipy as sp
import torch
import tensorflow as tf
import transformers
from transformers import pipeline
from transformers import RobertaTokenizer, RobertaModel
from transformers import AutoModelForSequenceClassification
from transformers import TFAutoModelForSequenceClassification
from transformers import AutoTokenizer
# from transformers_interpret import SequenceClassificationExplainer

device = "cuda:0" if torch.cuda.is_available() else "cpu"

tokenizer = AutoTokenizer.from_pretrained("paragon-analytics/ADRv1")  
model = AutoModelForSequenceClassification.from_pretrained("paragon-analytics/ADRv1").to(device)

# build a pipeline object to do predictions
pred = transformers.pipeline("text-classification", model=model, 
                             tokenizer=tokenizer, return_all_scores=True)

explainer = shap.Explainer(pred)

def interpretation_function(text):
    shap_values = explainer([text])
    scores = list(zip(shap_values.data[0], shap_values.values[0, :, 1]))
    return scores

# model = AutoModelForSequenceClassification.from_pretrained("paragon-analytics/ADRv1")
# modelc = AutoModelForSequenceClassification.from_pretrained("paragon-analytics/ADRv1").cuda


# cls_explainer = SequenceClassificationExplainer(
#     model,
#     tokenizer)

# # define a prediction function
# def f(x):
#     tv = torch.tensor([tokenizer.encode(v, padding='max_length', max_length=500, truncation=True) for v in x]).cuda()
#     outputs = modelc(tv)[0].detach().cpu().numpy()
#     scores = (np.exp(outputs).T / np.exp(outputs).sum(-1)).T
#     val = sp.special.logit(scores[:,1]) # use one vs rest logit units
#     return val

def adr_predict(x):
    encoded_input = tokenizer(str(x), return_tensors='pt')
    output = model(**encoded_input)
    scores = output[0][0].detach().numpy()
    scores = tf.nn.softmax(scores)
    
    # # build a pipeline object to do predictions
    # pred = transformers.pipeline("text-classification", model=model, 
    #                          tokenizer=tokenizer, device=0, return_all_scores=True)
    # explainer = shap.Explainer(pred)
    # shap_values = explainer([x])
    # shap_plot = shap.plots.text(shap_values)

    # word_attributions = cls_explainer(str(x))
    # # scores = list(zip(shap_values.data[0], shap_values.values[0, :, 1]))
    # 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))]

    # # SHAP:
    # # build an explainer using a token masker
    # explainer = shap.Explainer(f, tokenizer)
    # shap_values = explainer(str(x), fixed_context=1)
    # scores = list(zip(shap_values.data[0], shap_values.values[0, :, 1]))
    # # plot the first sentence's explanation
    # # plt = shap.plots.text(shap_values[0],display=False)
    shap_scores = interpretation_function(str(x).lower())


    return {"Severe Reaction": float(scores.numpy()[1]), "Non-severe Reaction": float(scores.numpy()[0])}, shap_scores
    # , word_attributions ,scores

def main(prob1):
    text = str(prob1).lower()
    obj = adr_predict(text)
    return obj[0],obj[1]
    # ,obj[2]

title = "Welcome to **ADR Detector** 🪐"
description1 = """
This app takes text (up to a few sentences) and predicts to what extent the text describes severe (or non-severe) adverse reaction to medicaitons.
"""

with gr.Blocks(title=title) as demo:
    gr.Markdown(f"## {title}")
    gr.Markdown(description1)
    gr.Markdown("""---""")
    prob1 = gr.Textbox(label="Enter Your Text Here:",lines=2, placeholder="Type it here ...")
    submit_btn = gr.Button("Analyze")


    with gr.Column(visible=True) as output_col:
        label = gr.Label(label = "Predicted Label")
        # impplot = gr.HighlightedText(label="Important Words", combine_adjacent=False).style(
        # color_map={"+++": "royalblue","++": "cornflowerblue",
        #  "+": "lightsteelblue", "NA":"white"})
        # NER = gr.HTML(label = 'NER:')
        # intp = gr.HighlightedText(label="Word Scores",
        # combine_adjacent=False).style(color_map={"++": "darkred","+": "red", 
        #                                         "--": "darkblue",
        #                                         "-": "blue", "NA":"white"})

        interpretation = gr.components.Interpretation(text)


    submit_btn.click(
        main,
        [prob1],
        [label
         # ,intp
         ,interpretation
        ], api_name="adr"
    )

    gr.Markdown("### Click on any of the examples below to see to what extent they contain resilience messaging:")
    gr.Examples([["I have minor pain."],["I have severe pain."]], [prob1], [label
                                                                           # ,intp
                                                                           ,interpretation
                                                                          ], main, cache_examples=True)
    
demo.launch()