main.py

import os
from flask import Flask, request, render_template
import pandas as pd
import torch
from transformers import BertTokenizer, BertForSequenceClassification
from collections import Counter
import matplotlib
matplotlib.use('Agg')  # Prevents GUI issues for Matplotlib
import matplotlib.pyplot as plt
import base64
from io import BytesIO

# Set writable cache directories within /tmp
os.environ["HF_HOME"] = "/tmp/huggingface_cache"  # Replaces TRANSFORMERS_CACHE
os.environ["MPLCONFIGDIR"] = "/tmp/matplotlib"

# Create directories if they don't exist
os.makedirs(os.environ["HF_HOME"], exist_ok=True)
os.makedirs(os.environ["MPLCONFIGDIR"], exist_ok=True)

app = Flask(__name__)

# Load Model from Local Directory
MODEL_PATH = "bert_imdb_model.bin"
TOKENIZER_PATH = "bert-base-uncased"

if os.path.exists(MODEL_PATH):
    print("Loading model from local file...")
    model = BertForSequenceClassification.from_pretrained('bert-base-uncased', num_labels=2)
    model.load_state_dict(torch.load(MODEL_PATH, map_location=torch.device('cpu')))
else:
    print(f"Error: Model file {MODEL_PATH} not found.")
    exit(1)

model.eval()
tokenizer = BertTokenizer.from_pretrained(TOKENIZER_PATH)

# ... rest of your code (keep the rest unchanged) ...


def predict_sentiment(text):
    tokens = tokenizer.encode(text, add_special_tokens=True)
    chunks = [tokens[i:i + 512] for i in range(0, len(tokens), 512)]

    sentiments = []
    for chunk in chunks:
        inputs = tokenizer.decode(chunk, skip_special_tokens=True, clean_up_tokenization_spaces=True)
        inputs = tokenizer(inputs, return_tensors="pt", truncation=True, padding=True, max_length=512)
        
        with torch.no_grad():
            outputs = model(**inputs)
        
        sentiments.append(outputs.logits.argmax(dim=1).item())

    majority_sentiment = Counter(sentiments).most_common(1)[0][0]
    return 'Positive' if majority_sentiment == 1 else 'Negative'

@app.route('/')
def upload_file():
    return render_template('upload.html')

@app.route('/analyze_text', methods=['POST'])
def analyze_text():
    text = request.form['text']
    sentiment = predict_sentiment(text)
    return render_template('upload.html', sentiment=sentiment)

@app.route('/uploader', methods=['GET', 'POST'])
def upload_file_post():
    if request.method == 'POST':
        f = request.files['file']
        data = pd.read_csv(f)

        data['sentiment'] = data['review'].apply(predict_sentiment)

        sentiment_counts = data['sentiment'].value_counts().to_dict()
        summary = f"Total Reviews: {len(data)}<br>" \
                  f"Positive: {sentiment_counts.get('Positive', 0)}<br>" \
                  f"Negative: {sentiment_counts.get('Negative', 0)}<br>"

        fig, ax = plt.subplots()
        ax.bar(sentiment_counts.keys(), sentiment_counts.values(), color=['red', 'blue'])
        ax.set_ylabel('Counts')
        ax.set_title('Sentiment Analysis Summary')

        img = BytesIO()
        plt.savefig(img, format='png', bbox_inches='tight')
        img.seek(0)
        plot_url = base64.b64encode(img.getvalue()).decode('utf8')
        plt.close(fig)

        return render_template('result.html', tables=[data.to_html(classes='data')], titles=data.columns.values, summary=summary, plot_url=plot_url)

if __name__ == '__main__':
    app.run(host='0.0.0.0', port=7860, debug=True)