app.py

import pandas as pd
import gradio as gr
from transformers import GPT2Tokenizer, GPT2LMHeadModel

# Load the model and tokenizer for GPT-2
tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
model = GPT2LMHeadModel.from_pretrained('gpt2')

df = pd.read_csv('anomalies.csv')
df['Feedback'] = ""  # Initialize with empty strings instead of None for consistency

# Preprocessing steps
df['ds'] = pd.to_datetime(df['ds']).dt.strftime('%Y-%m-%d')  # Format the datetime values
df['real'] = df['real'].apply(lambda x: f"{x:.2f}")  # Format the float values to two decimal places

# Convert each row into a structured natural language sentence
def tokenize_row(row):
    return f"On {row['ds']}, the expense in the group '{row['Group']}' was ${row['real']}."

# Apply the tokenization function to each row
df['tokenized'] = df.apply(tokenize_row, axis=1)

# Function to respond to questions with GPT-2
def answer_question_with_gpt(question):
    if tokenizer.pad_token is None:
        tokenizer.pad_token = tokenizer.eos_token

    # Simplify the prompt to avoid exceeding token limits
    latest_entries = df['tokenized'].tail(10).tolist()  # Limit to the last 10 entries for context
    prompt = f"Based on the following data: {' '.join(latest_entries)} Question: {question} Answer:"
    inputs = tokenizer(prompt, return_tensors='pt', padding='max_length', truncation=True, max_length=512)
    attention_mask = inputs['attention_mask']
    input_ids = inputs['input_ids']
    
    generated_ids = model.generate(
        input_ids,
        attention_mask=attention_mask,
        max_length=len(input_ids[0]) + 100,
        temperature=0.3,
        top_p=0.9,
        no_repeat_ngram_size=2
    )
    
    generated_text = tokenizer.decode(generated_ids[0], skip_special_tokens=True)
    # Extract the response after "Answer:"
    response_part = generated_text.split("Answer:")[1] if "Answer:" in generated_text else "No answer found."
    final_response = response_part.split(".")[0] + "."
    return final_response

# Function to add feedback
def add_feedback(name, feedback):
    global df
    if name in df['Group'].values:
        df.loc[df['Group'] == name, 'Feedback'] = feedback
        return "Feedback successfully added."
    else:
        return "Data not found in DataFrame."

with gr.Blocks() as demo:
    gr.Markdown("# Data Query and Feedback System")
    with gr.Row():
        with gr.Column():
            question_input = gr.Textbox(label="Ask a Question")
            answer_output = gr.Textbox(label="Answer", interactive=False)
            ask_button = gr.Button("Ask")
        with gr.Column():
            name_input = gr.Textbox(label="Name for Feedback")
            feedback_input = gr.Textbox(label="Feedback")
            feedback_result = gr.Textbox(label="Feedback Result", interactive=False)
            submit_button = gr.Button("Submit Feedback")

    ask_button.click(fn=answer_question_with_gpt, inputs=question_input, outputs=answer_output)
    submit_button.click(fn=add_feedback, inputs=[name_input, feedback_input], outputs=feedback_result)

demo.launch()