D3V1L1810/Multi_Text_Classification

1import tensorflow as tf2from transformers import BertTokenizer, TFBertForSequenceClassification3import numpy as np4import json5import requests6import gradio as gr7import logging8 9# Initialize the tokenizer and model10bert_tokenizer = BertTokenizer.from_pretrained('MultiTokenizer_ep10')11bert_model = TFBertForSequenceClassification.from_pretrained('MultiModel_ep10')12 13# Function to send results to API14# def send_results_to_api(data, result_url):15#     headers = {'Content-Type':'application/json'}16#     response = requests.post(result_url, json = data, headers=headers)17#     if response.status_code == 200:18#         return response.json19#     else:20#         return {'error':f"failed to send result to API: {response.status_code}"}21 22def predict_text(params):23    try:24        params = json.loads(params)25    except json.JSONDecodeError as e:26        logging.error(f"Invalid JSON input: {e.msg} at line {e.lineno} column {e.colno}")27        return {"error": f"Invalid JSON input: {e.msg} at line {e.lineno} column {e.colno}"}28        29    texts = params.get("urls", [])30    if not params.get("normalfileID", []):31        file_ids = [None] * len(texts)32    else:33        file_ids = params.get("normalfileID", [])34 35    if not texts:36        return {"error": "Missing required parameters: 'texts'"}37 38    solutions = []39    confidence_threshold = 0.85  # Define your confidence threshold40 41    for text, file_id in zip(texts, file_ids):42        encoding = bert_tokenizer.encode_plus(43            text,44            add_special_tokens=True,45            max_length=128,46            return_token_type_ids=True,47            padding='max_length',48            truncation=True,49            return_attention_mask=True,50            return_tensors='tf'51        )52        input_ids = encoding['input_ids']53        token_type_ids = encoding['token_type_ids']54        attention_mask = encoding['attention_mask']55 56        pred = bert_model.predict([input_ids, token_type_ids, attention_mask])57        logits = pred.logits58        softmax_scores = tf.nn.softmax(logits, axis=1).numpy()[0]59        pred_label = tf.argmax(logits, axis=1).numpy()[0]60 61        # Get the confidence score for the predicted label62        confidence = softmax_scores[pred_label]63        print(confidence)64        # If confidence is below the threshold, set answer to None65        if confidence < confidence_threshold:66            pred_label = 7  # Set to 'None' class67 68        label = {0: 'BUSINESS', 1: 'COMEDY', 2: 'CRIME', 3: 'FOOD & DRINK', 4: 'POLITICS', 5: 'SPORTS', 6: 'TRAVEL', 7: 'None'}69        result = {'text': text, 'answer': [label[pred_label]], "qcUser": None, "normalfileID": file_id}70        solutions.append(result)71 72    # result_url = f"{api}/{job_id}"73    # send_results_to_api(solutions, result_url)74    return json.dumps({"solutions": solutions})75 76inputt = gr.Textbox(label="Parameters in Json Format... Eg. {'texts':['text1', 'text2']}")77outputt = gr.JSON()78 79application = gr.Interface(fn=predict_text, inputs=inputt, outputs=outputt, title='Multi Text Classification with API Integration..')80application.launch()81