app.py

import os
import gradio as gr
import numpy as np
import torch
from transformers import pipeline
from sklearn.datasets import fetch_california_housing
from sklearn.linear_model import LinearRegression, LogisticRegression
from sklearn.model_selection import train_test_split
from sklearn.metrics import r2_score, accuracy_score

# ─── 1) LOAD A PUBLIC LLM ──────────────────────────────────────────────────────
device = 0 if torch.cuda.is_available() else -1
generator = pipeline(
    "text-generation",
    model="distilgpt2",
    tokenizer="distilgpt2",
    device=device,
    max_new_tokens=100,
    do_sample=True,
    temperature=0.7,
)

def chat_response(prompt: str) -> str:
    out = generator(prompt)
    return out[0]["generated_text"].strip()

# ─── 2) PREPARE HOUSING ML MODELS ───────────────────────────────────────────────
data = fetch_california_housing()
X, y = data.data, data.target

# linear regression
Xtr, Xte, ytr, yte = train_test_split(X, y, test_size=0.2, random_state=42)
reg = LinearRegression().fit(Xtr, ytr)
r2  = r2_score(yte, reg.predict(Xte))

# logistic classification
y_bin = (y > np.median(y)).astype(int)
Xtr2, Xte2, ytr2, yte2 = train_test_split(X, y_bin, test_size=0.2, random_state=42)
clf = LogisticRegression(max_iter=1000).fit(Xtr2, ytr2)
acc = accuracy_score(yte2, clf.predict(Xte2))

# ─── 3) ORCHESTRATOR / STATE MACHINE ─────────────────────────────────────────
def respond(message, chat_history, state):
    stage   = state.get("stage", "chat")
    ml_type = state.get("ml_type")

    # Clarify step
    if stage == "clarify":
        m = message.lower()
        if "linear" in m:
            state.update(stage="ml", ml_type="regression")
            chat_history.append(("Bot","Running linear regression…"))
        elif "class" in m:
            state.update(stage="ml", ml_type="classification")
            chat_history.append(("Bot","Running logistic classification…"))
        else:
            chat_history.append(("Bot","⚠️ Say ‘linear regression’ or ‘classification’."))
            return chat_history, state

    # ML execution
    if state["stage"] == "ml":
        if state["ml_type"] == "regression":
            chat_history.append(("Bot",f"✅ R²={r2:.3f}\nCoefs={np.round(reg.coef_,3).tolist()}"))
        else:
            chat_history.append(("Bot",f"✅ Accuracy={acc:.3f}"))
        state.update(stage="chat", ml_type=None)
        return chat_history, state

    # Detect intent
    if any(k in message.lower() for k in ["predict","regression","classification"]):
        state["stage"] = "clarify"
        chat_history.append(("Bot","Sure—linear regression or classification?"))
        return chat_history, state

    # Fallback to LLM
    reply = chat_response(message)
    chat_history.append(("Bot", reply))
    return chat_history, state

# ─── 4) GRADIO UI ───────────────────────────────────────────────────────────────
with gr.Blocks() as demo:
    gr.Markdown("## 🤖 LLM + Housing‐ML Chatbot")
    chat    = gr.Chatbot()
    user_in = gr.Textbox(placeholder="Type here…", show_label=False)
    state   = gr.State({"stage":"chat","ml_type":None})
    user_in.submit(respond, [user_in, chat, state], [chat, state])

# queue() creates the /api/predict endpoint that Spaces needs
demo = demo.queue()

if __name__ == "__main__":
    port = int(os.environ.get("PORT", 7860))
    demo.launch(server_name="0.0.0.0", server_port=port)