import gradio as gr
import pandas as pd
import matplotlib.pyplot as plt
import requests
import io
import os

# Token vem dos "Repository secrets" no Hugging Face
HF_TOKEN = os.getenv("HF_TOKEN")
MODEL = "google/timesfm-2.5-200m-pytorch"

API_URL = f"https://api-inference.huggingface.co/models/{MODEL}"
headers = {"Authorization": f"Bearer {HF_TOKEN}"}

def forecast(file, date_col, value_col, steps):
    # Lê CSV ou Excel
    if file.name.endswith(".csv"):
        df = pd.read_csv(file.name)
    else:
        df = pd.read_excel(file.name)

    # Converte coluna de datas
    df[date_col] = pd.to_datetime(df[date_col])
    df = df.sort_values(by=date_col)

    series = df[value_col].tolist()

    # Payload para a API
    payload = {
        "inputs": series,
        "parameters": {"prediction_length": steps}
    }

    response = requests.post(API_URL, headers=headers, json=payload)

    if response.status_code != 200:
        return f"Erro na API: {response.text}", None

    preds = response.json().get("prediction", series[-steps:])

    # Gráfico
    fig, ax = plt.subplots()
    ax.plot(df[date_col], df[value_col], label="Histórico")
    future_dates = pd.date_range(start=df[date_col].iloc[-1], periods=steps+1, freq="D")[1:]
    ax.plot(future_dates, preds, label="Previsão", linestyle="--")
    ax.legend()
    plt.title("📊 Previsão de Vendas (TimesFM)")

    buf = io.BytesIO()
    plt.savefig(buf, format="png")
    buf.seek(0)

    return "✅ Previsão concluída!", buf

with gr.Blocks() as demo:
    gr.Markdown("## 📈 Previsão de Vendas com TimesFM (Hugging Face)")

    file = gr.File(label="Envie seu arquivo (.csv ou .xlsx)", file_types=[".csv", ".xlsx"])
    date_col = gr.Textbox(label="Nome da coluna de datas")
    value_col = gr.Textbox(label="Nome da coluna de valores")
    steps = gr.Slider(1, 90, value=30, label="Quantos dias prever?")
    output_text = gr.Textbox(label="Resultado")
    output_plot = gr.Image(type="pil", label="Gráfico")

    btn = gr.Button("Gerar Previsão")
    btn.click(forecast, inputs=[file, date_col, value_col, steps], outputs=[output_text, output_plot])

demo.launch()