app.py

import gradio as gr
import numpy as np
import pandas as pd
import tensorflow as tf
import pickle
import spacy
from tqdm import tqdm
import gc
import os
import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification
from fastai.vision.all import *
from fastai.text.all import *
from torch.utils.data import Dataset
from DeBERTaV3 import ModelLoader

model_lst = ["DeBERTaV3", "BiLSTM"]

# BiLSTM Model
## Download the SpaCy model
os.system("python -m spacy download en_core_web_lg")

## Load models
model_1 = tf.keras.models.load_model("BiLSTM/model_1.h5")
model_2 = tf.keras.models.load_model("BiLSTM/model_2.h5")
model_3 = tf.keras.models.load_model("BiLSTM/model_3.h5")
model_4 = tf.keras.models.load_model("BiLSTM/model_4.h5")

## Load dictionaries
with open('BiLSTM/word_dict.pkl', 'rb') as f:
    word_dict = pickle.load(f)

## Load SpaCy NLP model
nlp = spacy.load('en_core_web_lg', disable=['parser', 'ner', 'tagger'])
nlp.vocab.add_flag(lambda s: s.lower() in spacy.lang.en.stop_words.STOP_WORDS, spacy.attrs.IS_STOP)

## tokenizer 
def preprocess_text(text):
    """Preprocess the input text using SpaCy and return word indices."""
    docs = nlp.pipe([text], n_process=1)
    word_seq = []
    for doc in docs:
        for token in doc:
            if token.pos_ != "PUNCT":
                if token.text not in word_dict:
                    word_dict[token.text] = 0 # OOV_INDEX
                word_seq.append(word_dict[token.text])
    return word_seq

# DeBERTaV3 Model
## Load tokenizer and model
tokenizer = AutoTokenizer.from_pretrained("microsoft/deberta-v3-base")

class QuestionDataset(Dataset):
    def __init__(self, X, y, tokenizer):
        self.text = X
        self.targets = y
        self.tok = tokenizer
    
    def __len__(self):
        return len(self.text)
    
    def __getitem__(self, idx):
        
        text = self.text[idx]
        targ = self.targets[idx]
        
        return self.tok(text, padding='max_length', 
                        truncation=True,
                        max_length=30,
                        return_tensors="pt")["input_ids"][0], tensor(targ)
    
    def new_empty(self):
        return QuestionDataset([], [], self.tok)
    
model_loader = ModelLoader()
learner = model_loader.get_learner()
print("Learner loaded successfully.")

## DataLoader
class TestDS:
    def __init__(self, tensors):
        self.tensors = tensors
    
    def __len__(self):
        return len(self.tensors)
    
    def __getitem__(self, idx):
        t = self.tensors[idx]
        return t, tensor(0)

class DeBERTaV3Model:
    def __init__(self):
        pass

    def predict(self, text):
        # Preprocess the text
        test_tensor = tokenizer(text, padding="max_length", truncation=True, max_length=55, return_tensors="pt")["input_ids"]
        test_dl = DataLoader(TestDS(test_tensor), bs=128)

        # Get predictions
        preds = learner.get_preds(dl=test_dl)
        label = "Insincere" if (F.softmax(preds[0], dim=1)[:, 1]>0.4878) else "Sincere"
        probs = {
            "Probability": float(F.softmax(preds[0], dim=1)[:, 1]),
            "Sequence": test_tensor[test_tensor != 0],
            "Decoded Sequence": tokenizer.decode(test_tensor[test_tensor != 0], skip_special_tokens=True)
        }
        return label, probs

class BiLSTMModel:
    def __init__(self):
        pass

    def predict(self, text):
        # Preprocess the text
        seq = preprocess_text(text)
        padded_seq = tf.keras.preprocessing.sequence.pad_sequences([seq], maxlen=55)

        BATCH_SIZE = 512
        # Get predictions from each model
        pred1 = 0.15 * np.squeeze(model_1.predict(padded_seq, batch_size=BATCH_SIZE, verbose=2))
        pred2 = 0.35 * np.squeeze(model_2.predict(padded_seq, batch_size=BATCH_SIZE, verbose=2))
        pred3 = 0.15 * np.squeeze(model_3.predict(padded_seq, batch_size=BATCH_SIZE, verbose=2))
        pred4 = 0.35 * np.squeeze(model_4.predict(padded_seq, batch_size=BATCH_SIZE, verbose=2))

        # Combine predictions
        avg_pred = pred1 + pred2 + pred3 + pred4
        label = "Insincere" if avg_pred > 0.35 else "Sincere"

        probs = {
            "Probability": float(avg_pred),
            "Model Probabilities": {"Model 1": float(pred1), "Model 2": float(pred2), "Model 3": float(pred3), "Model 4": float(pred4)},
            "Sequence": seq,
            "Decoded Sequence": " ".join([list(word_dict.keys())[list(word_dict.values()).index(i)] for i in seq])
        }
        return label, probs
    
class QuestionClassifier:
    """Main Class to manage the models"""
    def __init__(self):
        self.models = {
            "DeBERTaV3": DeBERTaV3Model(),
            "BiLSTM": BiLSTMModel()
        }

    def classify(self, model_name, text):
        return self.models[model_name].predict(text)
    
# Example questions
examples = [
    ["DeBERTaV3", "How do you train a pigeon to send messages?"],
    ["DeBERTaV3", "Is USA a shithole country owing to a shithole president?"],
    ["DeBERTaV3", "Why is Indian education a total bullshit?"],
    ["DeBERTaV3", "Which person has given the least f**ks and still turned out successful?"],
    ["BiLSTM", "How do you train a pigeon to send messages?"],
    ["BiLSTM", "Is USA a shithole country owing to a shithole president?"],
    ["BiLSTM", "Why is Indian education a total bullshit?"],
    ["BiLSTM", "Which person has given the least f**ks and still turned out successful?"]
]

def create_gradio_interface():
    classifier = QuestionClassifier()

    def classify_question(model_name, text):
        return classifier.classify(model_name, text)

    interface = gr.Interface(
        fn=classify_question,
        inputs=[
            gr.Dropdown(choices=["DeBERTaV3", "BiLSTM"], label="Select Model", value="BiLSTM"),
            gr.Textbox(lines=2, placeholder="Enter your question here...", label="Input Question")
        ],
        outputs=[
            gr.Textbox(label="Prediction"),
            gr.JSON(label="Model Probabilities")
        ],
        title="Quora Insincere Questions Classifier",
        examples=examples,
        description="Enter your question to classify it as sincere or insincere. Select an example question below."
    )
    interface.launch()


if __name__ == "__main__":
    create_gradio_interface()