Upload 2 files

app.py

@@ -0,0 +1,104 @@
+import os
+os.environ["HF_HOME"] = "/tmp/.cache/huggingface"
+os.environ["MPLCONFIGDIR"] = "/tmp/.config/matplotlib"
+
+import streamlit as st
+import pandas as pd
+import numpy as np
+import shap
+import joblib
+import matplotlib.pyplot as plt
+import matplotlib.font_manager as fm
+from huggingface_hub import hf_hub_download
+
+# 🚩 載入模型
+@st.cache_resource(show_spinner=True)
+def load_model():
+    model_path = hf_hub_download(
+        repo_id="jung-ming/god-eye-traffic-predictor",
+        filename="multi_output_lgbm.pkl",
+        repo_type="model"
+    )
+    return joblib.load(model_path)
+
+model = load_model()
+
+# 🎯 定義 SHAP 解釋器（只針對預測60分鐘的模型）
+explainer = shap.TreeExplainer(model.estimators_[0])
+
+# ⚙️ 載入字型（顯示中文）
+def find_chinese_font():
+    for font_path in fm.findSystemFonts(fontext='ttf'):
+        if "NotoSans" in font_path and ("TC" in font_path or "TraditionalChinese" in font_path):
+            return font_path
+        if "Heiti" in font_path or "LiHei" in font_path:
+            return font_path
+    return None
+
+font_path = find_chinese_font()
+if font_path:
+    chinese_font_prop = fm.FontProperties(fname=font_path)
+    plt.rcParams['font.family'] = chinese_font_prop.get_name()
+else:
+    plt.rcParams['font.family'] = 'DejaVu Sans'
+plt.rcParams['axes.unicode_minus'] = False
+
+# 🔢 Streamlit UI
+st.title("🚦 上帝視角：AI 國道壅塞預測系統")
+st.markdown("輸入當前交通特徵，預測 60–90 分鐘後之速率與壅塞風險。")
+
+# ✏️ 使用者輸入
+hour = st.selectbox("小時 (hour)", list(range(0, 24)))
+minute = st.selectbox("分鐘 (minute)", list(range(0, 60)))
+count_total = st.number_input("總車流量 (count_total)", 0, 5000, 1000)
+median_speed_avg = st.number_input("平均速率 (median_speed_avg)", 0.0, 200.0, 90.0)
+median_time_avg = st.number_input("平均時間 (median_time_avg)", 0.0, 1000.0, 600.0)
+start_detector_encode = st.number_input("起點檢測器編碼 (start_detector_encode)", 0.0, 400.0, 90.0)
+direction = st.selectbox("方向 (direction)", [0, 1], format_func=lambda x: "南向" if x == 0 else "北向")
+diff_speed_0_10 = st.number_input("0-10分鐘差速 (diff_speed_0_10)", -100.0, 100.0, 0.0)
+
+if st.button("🔮 預測壅塞狀況"):
+    input_data = pd.DataFrame({
+        'hour': [hour],
+        'minute': [minute],
+        'count_total': [count_total],
+        'median_speed_avg': [median_speed_avg],
+        'median_time_avg': [median_time_avg],
+        'start_detector_encode': [start_detector_encode],
+        'direction': [direction],
+        'diff_speed_0_10': [diff_speed_0_10]
+    })
+
+    y_pred = model.predict(input_data)[0]
+
+    def classify_risk(speed):
+        if speed < 40:
+            return "高風險"
+        elif speed < 60:
+            return "中風險"
+        else:
+            return "低風險"
+
+    risk_levels = [classify_risk(s) for s in y_pred]
+    st.subheader("📈 預測結果")
+    for i, t in enumerate(['60', '70', '80', '90']):
+        st.write(f"{t} 分鐘後速率：{y_pred[i]:.2f} km/h → 風險：{risk_levels[i]}")
+
+    st.subheader("🧠 模型決策解釋圖（60分鐘 SHAP Waterfall）")
+    shap_values = explainer.shap_values(input_data)
+    ax = shap.plots.waterfall(shap.Explanation(
+        values=shap_values[0],
+        base_values=explainer.expected_value,
+        data=input_data.iloc[0],
+        feature_names=input_data.columns
+    ), show=False)
+
+    # 設定中文字型（如果有）
+    if font_path:
+        for text in ax.texts:
+            text.set_fontproperties(chinese_font_prop)
+            if text.get_text().startswith('\u2212'):
+                text.set_text(text.get_text().replace('\u2212', '-'))
+
+    st.pyplot(ax.figure)
+    plt.close(ax.figure)

requirements.txt

@@ -0,0 +1,11 @@
+streamlit==1.35.0
+scikit-learn==1.3.2
+pandas==2.2.2
+numpy==1.24.4
+shap==0.45.0
+matplotlib==3.7.4
+joblib==1.3.2
+huggingface_hub==0.22.2
+folium==0.16.0
+geopandas==0.14.3
+fiona==1.9.6