app.py

import tensorflow as tf
from transformers import BertTokenizer, TFBertForSequenceClassification
import numpy as np
import json
import requests
import gradio as gr
import logging

# Initialize the tokenizer and model
bert_tokenizer = BertTokenizer.from_pretrained('MultiTokenizer_ep10')
bert_model = TFBertForSequenceClassification.from_pretrained('MultiModel_ep10')

# Function to send results to API
# def send_results_to_api(data, result_url):
#     headers = {'Content-Type':'application/json'}
#     response = requests.post(result_url, json = data, headers=headers)
#     if response.status_code == 200:
#         return response.json
#     else:
#         return {'error':f"failed to send result to API: {response.status_code}"}

def predict_text(params):
    try:
        params = json.loads(params)
    except json.JSONDecodeError as e:
        logging.error(f"Invalid JSON input: {e.msg} at line {e.lineno} column {e.colno}")
        return {"error": f"Invalid JSON input: {e.msg} at line {e.lineno} column {e.colno}"}
        
    texts = params.get("urls", [])
    if not params.get("normalfileID", []):
        file_ids = [None] * len(texts)
    else:
        file_ids = params.get("normalfileID", [])

    if not texts:
        return {"error": "Missing required parameters: 'texts'"}

    solutions = []
    confidence_threshold = 0.85  # Define your confidence threshold

    for text, file_id in zip(texts, file_ids):
        encoding = bert_tokenizer.encode_plus(
            text,
            add_special_tokens=True,
            max_length=128,
            return_token_type_ids=True,
            padding='max_length',
            truncation=True,
            return_attention_mask=True,
            return_tensors='tf'
        )
        input_ids = encoding['input_ids']
        token_type_ids = encoding['token_type_ids']
        attention_mask = encoding['attention_mask']

        pred = bert_model.predict([input_ids, token_type_ids, attention_mask])
        logits = pred.logits
        softmax_scores = tf.nn.softmax(logits, axis=1).numpy()[0]
        pred_label = tf.argmax(logits, axis=1).numpy()[0]

        # Get the confidence score for the predicted label
        confidence = softmax_scores[pred_label]
        print(confidence)
        # If confidence is below the threshold, set answer to None
        if confidence < confidence_threshold:
            pred_label = 7  # Set to 'None' class

        label = {0: 'BUSINESS', 1: 'COMEDY', 2: 'CRIME', 3: 'FOOD & DRINK', 4: 'POLITICS', 5: 'SPORTS', 6: 'TRAVEL', 7: 'None'}
        result = {'text': text, 'answer': [label[pred_label]], "qcUser": None, "normalfileID": file_id}
        solutions.append(result)

    # result_url = f"{api}/{job_id}"
    # send_results_to_api(solutions, result_url)
    return json.dumps({"solutions": solutions})

inputt = gr.Textbox(label="Parameters in Json Format... Eg. {'texts':['text1', 'text2']}")
outputt = gr.JSON()

application = gr.Interface(fn=predict_text, inputs=inputt, outputs=outputt, title='Multi Text Classification with API Integration..')
application.launch()