Sourav-003/StackExchange-Tag-Predictor

1 2import gradio as gr3import numpy as np4import pandas as pd5import re6from bs4 import BeautifulSoup7from tensorflow.keras.models import load_model8from tensorflow.keras.preprocessing.sequence import pad_sequences9from tensorflow.keras.preprocessing.text import Tokenizer10import tensorflow as tf11import pickle12import os # Import os for reading params.txt13 14x_voc_size = 12575 # Replace with your actual vocabulary size from your notebook15max_len = 100 # Replace with your actual max_len from your notebook16 17# Rebuild the model architecture18from tensorflow.keras.models import Sequential19from tensorflow.keras.layers import Embedding, GRU, Dense20model = Sequential()21model.add(Embedding(x_voc_size, 50, trainable = True, input_shape=(max_len,),mask_zero=True))22model.add(GRU(128))23model.add(Dense(128,activation='relu'))24model.add(Dense(10,activation='sigmoid'))25 26# Load the best weights27try:28    model.load_weights("weights.best.keras")29except Exception as e:30    print(f"Error loading model weights: {e}")31    print("Please ensure 'weights.best.keras' is in the same directory as app.py")32 33 34# Load the tokenizer35try:36    with open('x_tokenizer.pickle', 'rb') as handle:37        x_tokenizer = pickle.load(handle)38except FileNotFoundError:39    print("Error: x_tokenizer.pickle not found. Please ensure it is in the same directory as app.py")40    exit() # Exit if a critical file is missing41 42# Load the MultiLabelBinarizer43try:44    with open('mlb.pickle', 'rb') as handle:45        mlb = pickle.load(handle)46except FileNotFoundError:47    print("Error: mlb.pickle not found. Please ensure it is in the same directory as app.py")48    exit() # Exit if a critical file is missing49 50# Load opt and max_len from params.txt51opt = 0.34 # Default value, will try to load from file52max_len = 100 # Default value, will try to load from file53try:54    with open('params.txt', 'r') as f:55        for line in f:56            line = line.strip()57            if line.startswith("opt:"):58                opt = float(line.split(":")[1])59            elif line.startswith("max_len:"):60                max_len = int(line.split(":")[1])61except FileNotFoundError:62    print("Warning: params.txt not found. Using default values for opt and max_len.")63except Exception as e:64    print(f"Warning: Error reading params.txt: {e}. Using default values for opt and max_len.")65 66 67# Preprocessing function (same as in the notebook)68def cleaner(text):69  text = BeautifulSoup(text, "html.parser").get_text()70  text = re.sub("[^a-zA-Z]", " ", text)71  text = text.lower()72  tokens = text.split()73  return " ".join(tokens)74 75# Classification function (same as in the notebook)76def classify(pred_prob,thresh):77  y_pred_seq = []78  for i in pred_prob:79    temp=[]80    for j in i:81      if j>=thresh:82        temp.append(1)83      else:84        temp.append(0)85    y_pred_seq.append(temp)86  return y_pred_seq87 88# Prediction function (adapted for a single comment input)89def predict_tag(comment):90  #preprocess91  cleaned_text = [cleaner(comment)]92 93  #convert to integer sequences94  seq = x_tokenizer.texts_to_sequences(cleaned_text)95 96  #pad the sequence97  pad_seq = pad_sequences(seq,  padding='post', maxlen=max_len)98 99  #make predictions100  pred_prob = model.predict(pad_seq)101  classes = classify(pred_prob,opt)[0]102 103  classes = np.array([classes])104  predicted_tags = mlb.inverse_transform(classes)105  return ", ".join(predicted_tags[0]) # Return as a comma-separated string106 107# Create the Gradio interface108iface = gr.Interface(109    fn=predict_tag,110    inputs=gr.Textbox(lines=5, label="Enter StackExchange Question Text"),111    outputs=gr.Textbox(label="Predicted Tags"),112    title="StackExchange Tag Predictor",113    description="Predict tags for a given StackExchange question using a trained deep learning model."114)115 116# Launch the interface117if __name__ == "__main__":118    iface.launch(server_name="0.0.0.0", server_port=7860)119 120