# app.py
import gradio as gr
import joblib
import numpy as np
from pathlib import Path

# ---- find vectorizer in root or content ----
def find_file(name):
    for p in [Path("."), Path("content")]:
        f = p / name
        if f.exists():
            return f
    raise FileNotFoundError(f"Can't find {name} in repo root or ./content")

VECTORIZER = joblib.load(find_file("vectorizer.joblib"))

# ---- discover models (exclude vectorizer) in root + content ----
models = {}
for folder in [Path("."), Path("content")]:
    for p in folder.glob("*.joblib"):
        if p.name == "vectorizer.joblib":
            continue
        try:
            obj = joblib.load(p)
            if hasattr(obj, "predict"):
                models[p.stem] = obj
        except Exception:
            pass

if not models:
    raise RuntimeError("No models found. Place your *.joblib next to app.py or in ./content")

def predict(text, model_name):
    if not text.strip():
        return ""
    X = VECTORIZER.transform([text])
    y = int(models[model_name].predict(X)[0])
    return "Positive Feedback" if y == 1 else "Negative Feedback"

with gr.Blocks() as demo:
    gr.Markdown("# Sentiment Demo")
    with gr.Row():
        with gr.Column():
            txt = gr.Textbox(label="Review Comment", lines=6)
            mdl = gr.Dropdown(choices=sorted(models.keys()),
                              value=sorted(models.keys())[0],
                              label="method")
            btn = gr.Button("Submit")
        with gr.Column():
            out = gr.Textbox(label="Predicted Sentiment Class")
    btn.click(predict, [txt, mdl], out)

if __name__ == "__main__":
    demo.launch()