app.py

import gradio as gr
import pandas as pd
import joblib
from huggingface_hub import hf_hub_download

# =====================================================
# Load CICIDS2018 Model
# =====================================================
rf_model_path = hf_hub_download(
    repo_id="CodebaseAi/netraids-ml-models",
    filename="rf_pipeline.joblib"
)
rf_model = joblib.load(rf_model_path)

# =====================================================
# Load Training Artifacts (SAFE)
# =====================================================
artifacts = {}
try:
    artifacts_path = hf_hub_download(
        repo_id="CodebaseAi/netraids-ml-models",
        filename="training_artifacts.joblib"
    )
    artifacts = joblib.load(artifacts_path)
except Exception as e:
    print("Artifacts not loaded:", e)

# Try to infer feature columns safely
FEATURE_COLUMNS = None
for key in ["feature_columns", "features", "columns", "X_columns"]:
    if key in artifacts:
        FEATURE_COLUMNS = artifacts[key]
        break

# Try to infer class mapping safely
CLASS_MAPPING = None
for key in ["class_mapping", "label_mapping", "classes", "target_mapping"]:
    if key in artifacts:
        CLASS_MAPPING = artifacts[key]
        break

# =====================================================
# Load BCC-Darknet Model (5-class)
# =====================================================
darknet_model = joblib.load(
    hf_hub_download("CodebaseAi/netraids-ml-models", "realtime_model.pkl")
)
darknet_scaler = joblib.load(
    hf_hub_download("CodebaseAi/netraids-ml-models", "realtime_scaler.pkl")
)
darknet_encoder = joblib.load(
    hf_hub_download("CodebaseAi/netraids-ml-models", "realtime_encoder.pkl")
)

# =====================================================
# Prediction Router
# =====================================================
def predict(model_choice, features: dict):
    df = pd.DataFrame([features])

    if model_choice == "CICIDS2018 (13 Classes)":
        # Enforce feature order ONLY if available
        if FEATURE_COLUMNS is not None:
            df = df[FEATURE_COLUMNS]

        pred_idx = rf_model.predict(df)[0]

        if CLASS_MAPPING is not None:
            pred_label = CLASS_MAPPING.get(pred_idx, str(pred_idx))
        else:
            pred_label = str(pred_idx)

        return {
            "dataset": "CICIDS2018",
            "prediction": pred_label
        }

    else:
        X_scaled = darknet_scaler.transform(df)
        pred_encoded = darknet_model.predict(X_scaled)[0]
        pred_label = darknet_encoder.inverse_transform([pred_encoded])[0]

        return {
            "dataset": "BCC-Darknet",
            "prediction": str(pred_label)
        }

# =====================================================
# Gradio UI
# =====================================================
app = gr.Interface(
    fn=predict,
    inputs=[
        gr.Dropdown(
            choices=[
                "CICIDS2018 (13 Classes)",
                "BCC-Darknet (5 Classes)"
            ],
            label="Select Dataset"
        ),
        gr.JSON(label="Network Traffic Features (JSON)")
    ],
    outputs=gr.JSON(label="Detection Result"),
    title="NetraIDS – Dual-Model Network Intrusion Detection",
    description="Robust cloud-deployed ML inference with dataset-specific pipelines"
)

app.launch()