app.py

"""
Jade TrainerBox - API Gradio com ZeroGPU
Space isolado para treino de modelos ML.
"""
import gradio as gr

# Tentar importar spaces (só funciona no HF Spaces)
try:
    import spaces
    HAS_SPACES = True
except ImportError:
    HAS_SPACES = False
    # Mock decorator para rodar localmente
    class spaces:
        @staticmethod
        def GPU(duration=120):
            def decorator(func):
                return func
            return decorator

from trainer import train_model, run_eda


# Wrapper com ZeroGPU
@spaces.GPU(duration=120)
def train_with_gpu(csv_data: str, target_col: str, model_type: str) -> dict:
    """
    Treina modelo usando ZeroGPU (quando disponível).
    """
    return train_model(csv_data, target_col, model_type)


def api_train(csv_data: str, target_col: str, model_type: str = "xgboost") -> dict:
    """
    Endpoint principal de treino.
    
    Args:
        csv_data: CSV como string
        target_col: Nome da coluna target
        model_type: "xgboost", "lightgbm", ou "mlp"
    """
    if not csv_data or not csv_data.strip():
        return {"success": False, "error": "CSV vazio"}
    
    if not target_col or not target_col.strip():
        return {"success": False, "error": "Coluna target não especificada"}
    
    # Usar GPU se disponível (MLP se beneficia mais)
    if HAS_SPACES and model_type == "mlp":
        return train_with_gpu(csv_data, target_col, model_type)
    else:
        return train_model(csv_data, target_col, model_type)


def api_eda(csv_data: str) -> dict:
    """
    Endpoint de análise exploratória.
    """
    if not csv_data or not csv_data.strip():
        return {"success": False, "error": "CSV vazio"}
    
    return run_eda(csv_data)


# Interface Gradio
with gr.Blocks(title="Jade TrainerBox 🧠") as demo:
    gr.Markdown("# 🧠 Jade TrainerBox")
    gr.Markdown("Space isolado para treino de modelos ML. Use a API programaticamente.")
    
    with gr.Tab("Treino"):
        with gr.Row():
            csv_input = gr.Textbox(
                label="CSV Data",
                placeholder="col1,col2,target\n1,2,0\n3,4,1",
                lines=5
            )
        with gr.Row():
            target_input = gr.Textbox(label="Coluna Target", placeholder="target")
            model_dropdown = gr.Dropdown(
                choices=["xgboost", "lightgbm", "mlp"],
                value="xgboost",
                label="Modelo"
            )
        train_btn = gr.Button("Treinar", variant="primary")
        train_output = gr.JSON(label="Resultado")
        
        train_btn.click(
            fn=api_train,
            inputs=[csv_input, target_input, model_dropdown],
            outputs=train_output,
            api_name="train"
        )
    
    with gr.Tab("EDA"):
        eda_csv = gr.Textbox(
            label="CSV Data",
            placeholder="Dados para análise exploratória",
            lines=5
        )
        eda_btn = gr.Button("Analisar", variant="secondary")
        eda_output = gr.JSON(label="Análise")
        
        eda_btn.click(
            fn=api_eda,
            inputs=eda_csv,
            outputs=eda_output,
            api_name="eda"
        )
    
    gr.Markdown("---")
    gr.Markdown("### 📡 API Usage")
    gr.Markdown("""
    ```python
    from gradio_client import Client
    
    client = Client("seu-usuario/jade-trainerbox")
    result = client.predict(
        csv_data="col1,col2,target\\n1,2,0\\n3,4,1",
        target_col="target",
        model_type="xgboost",
        api_name="/train"
    )
    print(result)
    ```
    """)

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