app.py

# app.py (robust, server-safe)
import gradio as gr
import pandas as pd
import joblib
import shap
import numpy as np
import matplotlib
matplotlib.use("Agg")  # 非交互后端，服务器端更稳
import matplotlib.pyplot as plt
import warnings
warnings.filterwarnings("ignore")

# ====== 模型与背景数据 ======
MODEL_PATH = "models/SVM_pipeline.pkl"
BG_PATH    = "data/bg.csv"

# 模型最终需要的10个特征（顺序必须与训练一致）
feature_names = ["HGB", "HDL_C", "DBIL", "AST_ALT", "UA", "GFR", "PNI", "HALP", "AAPR", "conuts"]

# 加载模型与背景
pipeline = joblib.load(MODEL_PATH)
bg_df = pd.read_csv(BG_PATH)
bg_array = bg_df[feature_names].to_numpy(dtype=np.float64)

# 预测函数（供 KernelExplainer 调用）
def _predict_proba_nd(x_nd: np.ndarray) -> np.ndarray:
    df = pd.DataFrame(x_nd, columns=feature_names)
    return pipeline.predict_proba(df)

# 只初始化一次 explainer（性能更稳）
explainer = shap.KernelExplainer(_predict_proba_nd, bg_array)

def _render_force_plot(base_val: float, shap_1d: np.ndarray, feat_1d: np.ndarray, fnames):
    """返回 matplotlib Figure（旧接口，服务器端稳定）"""
    plt.close('all')
    shap.force_plot(
        base_val, np.asarray(shap_1d).reshape(-1), np.asarray(feat_1d).reshape(-1),
        feature_names=list(fnames), matplotlib=True, show=False
    )
    fig = plt.gcf()
    fig.set_size_inches(8, 4)
    plt.tight_layout()
    return fig

def predict_and_explain(
    HGB, HDL_C, DBIL, AST_ALT, UA, GFR,
    ALB, LYM, PLT, ALP, CHOL,
    nsamples=200
):
    status = []
    try:
        # ---- 1) 衍生指标（由原始输入计算）----
        try:
            HGB  = float(HGB);   HDL_C = float(HDL_C); DBIL = float(DBIL); AST_ALT = float(AST_ALT)
            UA   = float(UA);    GFR   = float(GFR)
            ALB  = float(ALB);   LYM   = float(LYM);   PLT   = float(PLT)
            ALP  = float(ALP);   CHOL  = float(CHOL)
        except Exception:
            return None, None, "Error: some inputs are not numeric."

        # 防极端值（避免除0）
        if PLT <= 0 or ALP <= 0:
            return None, None, "Error: PLT and ALP must be > 0."

        PNI  = ALB + 5.0 * LYM
        HALP = HGB * ALB * LYM / PLT
        AAPR = ALB / ALP
        conuts = (
            (0 if ALB >= 35 else 2 if ALB >= 30 else 4 if ALB >= 25 else 6) +
            (0 if LYM >= 1.6 else 1 if LYM >= 1.2 else 2 if LYM >= 0.8 else 3) +
            (0 if CHOL >= 4.65 else 1 if CHOL >= 3.10 else 2 if CHOL >= 2.59 else 3)
        )
        x_row = np.array([[HGB, HDL_C, DBIL, AST_ALT, UA, GFR, PNI, HALP, AAPR, conuts]], dtype=np.float64)
        status.append(f"Derived: PNI={PNI:.3f}, HALP={HALP:.3f}, AAPR={AAPR:.3f}, CONUTS={conuts}")

        # ---- 2) 概率 ----
        prob = float(pipeline.predict_proba(pd.DataFrame(x_row, columns=feature_names))[0, 1])
        status.append(f"Pred prob computed: {prob:.3f}")

        # ---- 3) SHAP 计算 ----
        ns = int(nsamples) if nsamples is not None else 200
        shap_out = explainer.shap_values(x_row, nsamples=ns)

        # 统一提取“正类”一维向量
        if isinstance(shap_out, list):
            sv = np.asarray(shap_out[1], dtype=np.float64)
            if sv.ndim == 2:
                sv = sv[0, :]
        else:
            sv = np.asarray(shap_out, dtype=np.float64)
            if sv.ndim == 3:      # (1, n_features, n_classes)
                sv = sv[0, :, 1]
            elif sv.ndim == 2:    # (1, n_features)
                sv = sv[0, :]
            else:
                sv = sv.reshape(-1)
        status.append(f"SHAP 1D shape: {sv.shape}; features: {x_row.shape[1:]}")

        # base value 取正类
        ev = explainer.expected_value
        if isinstance(ev, (list, np.ndarray)):
            ev = np.asarray(ev).reshape(-1)
            base_val = float(ev[1] if len(ev) > 1 else ev[0])
        else:
            base_val = float(ev)

        # ---- 4) 绘图：优先力图；失败则条形图兜底 ----
        try:
            fig = _render_force_plot(base_val, sv, x_row[0, :], feature_names)
            status.append("Rendered force plot (matplotlib).")
            return round(prob, 3), fig, "\n".join(status)
        except Exception as e_force:
            status.append(f"Force-plot failed: {repr(e_force)}; fallback=bar")

        order = np.argsort(np.abs(sv))[::-1]
        topk = order[:min(10, sv.shape[0])]
        plt.close('all')
        fig = plt.figure(figsize=(8, 5), dpi=160)
        plt.barh(np.array(feature_names)[topk], sv[topk])
        plt.xlabel("SHAP value")
        plt.title("Top features (single-sample contribution)")
        plt.gca().invert_yaxis()
        plt.tight_layout()
        status.append("Rendered bar fallback.")
        return round(prob, 3), fig, "\n".join(status)

    except Exception as e:
        return None, None, f"Fatal error: {repr(e)}"

# ====== 示例：一组“原始指标”可复现你之前的 PNI/HALP/AAPR/CONUTS ======
# 对应：PNI=44, HALP≈60.8, AAPR≈0.486, CONUTS=4
example_values = [167, 1.76, 8.6, 0.97, 310, 75, 33, 2.2, 164, 68, 2.8, 200]
# 顺序：HGB, HDL_C, DBIL, AST_ALT, UA, GFR, ALB, LYM, PLT, ALP, CHOL, nsamples

# ====== Gradio 界面 ======
with gr.Blocks() as demo:
    gr.Markdown(
        "### Meige Risk Prediction (SVM) with SHAP Explanation\n"
        "Enter **original clinical indicators**; the app will derive PNI/HALP/AAPR/CONUTS internally.\n\n"
        "**Units**: HGB (g/L), HDL‑C (mmol/L), DBIL (μmol/L), AST/ALT (ratio), UA (μmol/L), "
        "GFR (mL/min/1.73 m²), ALB (g/L), LYM (×10⁹/L), PLT (×10⁹/L), ALP (U/L), CHOL (mmol/L)."
    )

    with gr.Row():
        with gr.Column(scale=1):
            inputs = [
                gr.Number(label="HGB (g/L)"),
                gr.Number(label="HDL-C (mmol/L)"),
                gr.Number(label="DBIL (μmol/L)"),
                gr.Number(label="AST/ALT"),
                gr.Number(label="UA (μmol/L)"),
                gr.Number(label="GFR (mL/min/1.73 m²)"),
                gr.Number(label="ALB (g/L)"),
                gr.Number(label="LYM (×10⁹/L)"),
                gr.Number(label="PLT (×10⁹/L)"),
                gr.Number(label="ALP (U/L)"),
                gr.Number(label="CHOL (mmol/L)")
            ]
            ns_slider = gr.Slider(100, 400, value=200, step=50, label="SHAP nsamples")

            btn_fill = gr.Button("Fill Example")
            btn_predict = gr.Button("Predict")

        with gr.Column(scale=1):
            out_prob = gr.Number(label="Predicted Probability")
            out_plot = gr.Plot(label="SHAP Force Plot (fallback: bar)")
            out_log  = gr.Textbox(label="Status", lines=6)

    def _fill_example():
        return tuple(example_values)

    btn_fill.click(fn=_fill_example, outputs=[*inputs, ns_slider])
    btn_predict.click(
        fn=predict_and_explain,
        inputs=[*inputs, ns_slider],
        outputs=[out_prob, out_plot, out_log]
    )

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