deployed v1.0.0

app.py

@@ -1,57 +1,161 @@
+from huggingface_hub import hf_hub_download
 import streamlit as st
 import pandas as pd
-from huggingface_hub import hf_hub_download
 import joblib
-from sklearn.preprocessing import LabelEncoder
-from sklearn.metrics import accuracy_score
+import matplotlib.pyplot as plt
+import seaborn as sns
+from scipy.interpolate import make_interp_spline
+import numpy as np
 
-def load_model_from_hf_hub(repo_id, filename):
+def load_model(repo_id, filename):
     # 从Hugging Face Hub下载模型文件
     model_path = hf_hub_download(repo_id=repo_id, filename=filename)
-    # 加载模型
+    # 使用joblib加载模型
     model = joblib.load(model_path)
     return model
 
-def preprocess_data(uploaded_file, features):
-    # 读取并预处理新数据集
-    new_data = pd.read_csv(uploaded_file)
-    X_new = new_data[features]
-    return X_new, new_data['point_victor']
+# 设置Seaborn的风格
+sns.set(style="whitegrid")
 
-def main():
-    st.title('ML Model Prediction Demo with XGBoost')
+def preprocess_and_predict(model, data, match_id, features):
+    # 筛选出特定match_id的数据
+    specific_match_data = data[data['match_id'] == match_id]
+
+    if specific_match_data.empty:
+        st.write(f"No data found for match_id {match_id}.")
+        return None, None
+    else:
+        specific_match_data = specific_match_data.sort_values('elapsed_time')
+
+        # 预处理数据集
+        X_specific_match = specific_match_data[features]
+
+        # 使用模型进行概率预测
+        positive_class_probabilities = model.predict_proba(X_specific_match)[:, 1]
 
-    # 文件上传器
-    uploaded_file = st.file_uploader("Upload your input CSV file", type="csv")
+        return specific_match_data, positive_class_probabilities
+
+def plot_results(specific_match_data, positive_class_probabilities, match_id, show_true_label=True, show_predicted_probability=True, show_momentum=True):
+    plt.figure(figsize=(14, 7))
+    
+    if show_predicted_probability:
+        # 绘制预测概率的线图
+        sns.lineplot(
+            x=specific_match_data['elapsed_time'], 
+            y=positive_class_probabilities, 
+            marker='o', 
+            linestyle='-', 
+            color='blue', 
+            label='Predicted Probability'
+        )
     
+    if show_true_label:
+        # 绘制真实标签的散点图
+        sns.scatterplot(
+            x=specific_match_data['elapsed_time'], 
+            y=(specific_match_data['point_victor'] - 1), 
+            color='red', 
+            label='True Label', 
+            s=60
+        )
+
+    if show_momentum:
+        # 计算并绘制Momentum
+        window_size = 3
+        padded_probabilities = np.pad(positive_class_probabilities, (window_size//2, window_size//2), 'edge')
+        momentum = np.convolve(padded_probabilities, np.ones(window_size)/window_size, mode='same')[:len(specific_match_data)]
+        
+        # 创建平滑曲线
+        X_smooth = np.linspace(specific_match_data['elapsed_time'].min(), specific_match_data['elapsed_time'].max(), 300)
+        spline = make_interp_spline(specific_match_data['elapsed_time'], momentum, k=3)  # k是平滑曲线的强度
+        momentum_smooth = spline(X_smooth)
+        
+        # 绘制Momentum的平滑曲线下面积，半透明
+        plt.fill_between(X_smooth, momentum_smooth, color='green', alpha=0.3)
+        # 也可以选择绘制平滑曲线的边界，如果需要
+        plt.plot(X_smooth, momentum_smooth, color='green', label='Momentum')
+    
+    # 标记set_no和game_no变化的时刻
+    for i in range(1, len(specific_match_data)):
+        if specific_match_data['game_no'].iloc[i] != specific_match_data['game_no'].iloc[i-1]:
+            plt.axvline(
+                x=specific_match_data['elapsed_time'].iloc[i], 
+                color='gray', 
+                linestyle='--', 
+                lw=2
+            )
+        if specific_match_data['set_no'].iloc[i] != specific_match_data['set_no'].iloc[i-1]:
+            plt.axvline(
+                x=specific_match_data['elapsed_time'].iloc[i], 
+                color='red', 
+                linestyle='-.', 
+                lw=2
+            )
+    
+    plt.title(f'Predicted Probability, Momentum, and True Label Over Time for Match {match_id}')
+    plt.xlabel('Elapsed Time')
+    plt.ylabel('Probability / True Label / Momentum')
+    plt.grid(True)
+    plt.legend()
+    st.pyplot(plt)
+
+def main():
+    st.title('Momentum Catcher')
+
+    uploaded_file = st.file_uploader("Upload your input CSV data", type="csv")
+    match_id_input = st.text_input("Enter the match_id you want to analyze", "2023-wimbledon-1301")
+
     if uploaded_file is not None:
-        # 指定特征列
-        features = [
-            'elapsed_time', 'set_no', 'game_no', 'game_advantage', 'set_advantage', 
-            'p1_cumulative_distance', 'p2_cumulative_distance', 
-            'p1_max_continuous_scoring', 'p2_max_continuous_scoring', 
-            'p1_total_errors', 'p2_total_errors'
-        ]
-        # 指定Hugging Face Hub上模型的仓库ID和文件名
         repo_id = "Nagi-ovo/Momentum-XGboost"
         filename = "xgboost.pkl"
         
-        # 从Hugging Face Hub加载模型
-        xgb_model = load_model_from_hf_hub(repo_id, filename)
+        # 加载模型
+        xgb_model = load_model(repo_id, filename)
+        features = ['PAI_diff', 'normalized_rally', 'is_game_point']
         
-        # 预处理数据
-        X_new, y_new = preprocess_data(uploaded_file, features)
+        # 读取上传的CSV文件
+        new_data = pd.read_csv(uploaded_file)
+        new_data['is_game_point'] = abs(new_data['p1_facing_game_point'] - new_data['p2_facing_game_point'])
+        new_data['PAI_diff'] = new_data['p1_PAI'] - new_data['p2_PAI']
         
-        # 如果目标变量也需要编码转换，确保使用与训练数据相同的方式进行转换
-        label_encoder = LabelEncoder()
-        y_new_encoded = label_encoder.fit_transform(y_new)
-
-        # 使用加载的模型进行预测
-        predictions = xgb_model.predict(X_new)
-
-        # 计算并显示正确率
-        accuracy = accuracy_score(y_new_encoded, predictions)
-        st.write("Accuracy on new data:", accuracy)
-
+        specific_match_data, positive_class_probabilities = preprocess_and_predict(xgb_model, new_data, match_id_input, features)
+        
+        if specific_match_data is not None:
+            # 允许用户选择是否要观察特定的set和game
+            observe_specific = st.checkbox("Observe specific set(s) and game(s)", False)
+            
+            if observe_specific:
+                # 用户选择观察多盘
+                unique_sets = specific_match_data['set_no'].unique()
+                selected_sets = st.multiselect('Select set(s) to observe', unique_sets, default=unique_sets)
+                
+                # 基于选定的set_no，选择观察一局或多局
+                if selected_sets:
+                    filtered_data = specific_match_data[specific_match_data['set_no'].isin(selected_sets)]
+                    unique_games = filtered_data['game_no'].unique()
+                    selected_games = st.multiselect('Select games to observe (default is all games)', unique_games, default=unique_games)
+                    
+                    # 进一步筛选数据以仅包含选定的game_no
+                    if selected_games:
+                        filtered_data = filtered_data[filtered_data['game_no'].isin(selected_games)]
+                else:
+                    filtered_data = specific_match_data  # 如果没有选择任何set，显示全部数据
+            else:
+                filtered_data = specific_match_data
+                selected_games = specific_match_data['game_no'].unique()  # 默认选择所有游戏
+            
+            # 显示选项复选框
+            show_true_label = st.checkbox("Show True Label", True)
+            show_predicted_probability = st.checkbox("Show Predicted Probability", True)
+            show_momentum = st.checkbox("Show Momentum", True)
+            
+            # 如果筛选后的数据不为空，则绘制图表
+            if not filtered_data.empty:
+                # 计算筛选数据的预测概率
+                filtered_positive_class_probabilities = xgb_model.predict_proba(filtered_data[features])[:, 1]
+                plot_results(filtered_data, filtered_positive_class_probabilities, match_id_input, show_true_label, show_predicted_probability, show_momentum)
+            else:
+                st.write("No data available for the selected set(s) and games. Please select again.")
+            
 if __name__ == '__main__':
     main()