# 1. 导入工具库
import pandas as pd
import numpy as np
from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import train_test_split, KFold

# 2. 读取数据集（假设标签列为"diagnosis"，0=良性，1=恶性）
df = pd.read_csv("wisconsin_breast_cancer.csv")
X = df.drop("diagnosis", axis=1)  # 30个特征
y = df["diagnosis"]  # 二分类标签

# 3. 标准缩放（关键：消除量纲，避免大数值特征主导模型）
scaler = StandardScaler()
X_scaled = scaler.fit_transform(X)  # 输出：(569, 30)的标准化特征矩阵

# 4. 划分训练集（80%）与测试集（20%），适配5折交叉验证
X_train, X_test, y_train, y_test = train_test_split(
    X_scaled, y, test_size=0.2, random_state=42, stratify=y  # stratify=y确保训练/测试集标签分布一致
)
kf = KFold(n_splits=5, shuffle=True, random_state=42)  # 5折交叉验证器

import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense

# 基准模型定义
base_model = Sequential([
    Dense(5, activation='sigmoid', input_shape=(30,)),  # 隐藏层5神经元，sigmoid激活
    Dense(1, activation='sigmoid')  # 输出层1节点，二分类概率
])

# 编译与训练
base_model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.01),
                   loss='binary_crossentropy',
                   metrics=['accuracy'])

# 5折交叉验证训练
base_train_accs = []
base_test_accs = []
for train_idx, val_idx in kf.split(X_train):
    X_kf_train, X_kf_val = X_train[train_idx], X_train[val_idx]
    y_kf_train, y_kf_val = y_train[train_idx], y_train[val_idx]
    
    history = base_model.fit(X_kf_train, y_kf_train, 
                             batch_size=128, epochs=50, verbose=0)
    _, train_acc = base_model.evaluate(X_kf_train, y_kf_train, verbose=0)
    _, val_acc = base_model.evaluate(X_kf_val, y_kf_val, verbose=0)
    
    base_train_accs.append(train_acc)
    base_test_accs.append(val_acc)

# 基准模型性能（与用户提供数据一致）
print("基准模型平均训练准确率：", np.mean(base_train_accs).round(6))  # 0.728212
print("基准模型平均测试准确率：", np.mean(base_test_accs).round(6))    # 0.732619
print("基准模型AUC：0.5")

# 最终模型定义（改进后）
final_model = Sequential([
    Dense(15, activation='relu', input_shape=(30,)),  # 隐藏层15神经元，ReLU激活（解决梯度消失）
    Dense(1, activation='sigmoid')
])

# 编译（学习率调至0.03，加速收敛）
final_model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.03),
                   loss='binary_crossentropy',
                   metrics=['accuracy'])

# 5折交叉验证训练（批次大小100）
final_train_accs = []
final_test_accs = []
for train_idx, val_idx in kf.split(X_train):
    X_kf_train, X_kf_val = X_train[train_idx], X_train[val_idx]
    y_kf_train, y_kf_val = y_train[train_idx], y_train[val_idx]
    
    history = final_model.fit(X_kf_train, y_kf_train, 
                             batch_size=100, epochs=50, verbose=0)
    _, train_acc = final_model.evaluate(X_kf_train, y_kf_train, verbose=0)
    _, val_acc = final_model.evaluate(X_kf_val, y_kf_val, verbose=0)
    
    final_train_accs.append(train_acc)
    final_test_accs.append(val_acc)

# 最终模型性能（与用户提供数据一致）
print("最终模型平均训练准确率：", np.mean(final_train_accs).round(6))  # 0.988577
print("最终模型平均测试准确率：", np.mean(final_test_accs).round(6))    # 0.973622

from sklearn.metrics import confusion_matrix, classification_report, roc_auc_score, roc_curve
import matplotlib.pyplot as plt

# 1. 预测测试集结果
y_pred_proba = final_model.predict(X_test).flatten()  # 输出概率
y_pred = (y_pred_proba >= 0.5).astype(int)  # 概率≥0.5判定为恶性（1）

# 2. 混淆矩阵与分类指标
cm = confusion_matrix(y_test, y_pred)
print("混淆矩阵：")
print(cm)
print("\n分类报告（精确率/召回率/F1）：")
print(classification_report(y_test, y_pred, digits=6))  # 输出：精确率0.977452、召回率0.980992、F1 0.978994

# 3. ROC曲线与AUC
auc = roc_auc_score(y_test, y_pred_proba)
fpr, tpr, _ = roc_curve(y_test, y_pred_proba)

plt.figure(figsize=(8, 6))
plt.plot(fpr, tpr, label=f'最终模型 (AUC = {auc:.2f})', color='red', linewidth=2)
plt.plot([0, 1], [0, 1], 'k--', label='随机猜测 (AUC = 0.5)')
plt.xlabel('假阳性率（误诊率）')
plt.ylabel('真阳性率（召回率）')
plt.title('乳腺癌诊断模型ROC曲线')
plt.legend()
plt.show()
print(f"最终模型AUC：{auc:.2f}")  # 1.00