KerasTuner超参数优化中自定义指标（如F1、AUC）的正确配置方法

import keras_tuner as kt
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers
from tensorflow.keras.models import Sequential
from tensorflow.keras.optimizers import Adam
# 导入Keras内置的F1Score和AUC指标
from tensorflow.keras.metrics import F1Score, AUC 

class MyHyperModel(kt.HyperModel):
    def build(self, hp):
        model = Sequential()
        model.add(layers.Flatten())
        model.add(
            layers.Dense(
                units=hp.Int("units", min_value=24, max_value=128, step=10),
                activation="relu",
            )
        )
        model.add(layers.Dense(1, activation="sigmoid")) # 二分类输出

        # 编译模型时，明确指定要跟踪的指标
        # 关键点：使用Metric类实例，并可通过name参数指定其在日志中的名称
        # KerasTuner将会在日志中寻找 'val_f1_score' 和 'val_auc_score'
        model.compile(
            optimizer=Adam(learning_rate=hp.Float('learning_rate', 5e-5, 5e-1, step=0.001)),
            loss='binary_crossentropy',
            metrics=[
                F1Score(name='f1_score', threshold=0.5), # 为F1 Score指定日志名称
                AUC(name='auc_score')                   # 为AUC指定日志名称
            ]
        )
        return model

    def fit(self, hp, model, *args, **kwargs):
        return model.fit(
            *args,
            batch_size=hp.Choice("batch_size", [16, 32, 52]),
            epochs=hp.Int('epochs', min_value=5, max_value=25, step=5),
            **kwargs,
        )

# KerasTuner配置
tuner = kt.RandomSearch(
    MyHyperModel(),
    # 目标名称必须与模型编译时指标的验证集版本匹配
    # 如果编译时指标名称是 'f1_score'，则验证集目标是 'val_f1_score'
    # 如果要优化AUC，则 objective=kt.Objective("val_auc_score", direction="max")
    objective=kt.Objective("val_f1_score", direction="max"), # 优化F1 Score，方向为最大化
    max_trials=100,
    overwrite=True,
    directory="my_dir",
    project_name="tune_hypermodel_custom_metrics",
)

# 假设 X_train, y_train, X_test, y_test 已定义并加载
# tuner.search(X_train, y_train, validation_data=(X_test, y_test), callbacks=[keras.callbacks.EarlyStopping('val_loss', patience=3)])

print("KerasTuner配置完成，请准备数据并调用 tuner.search() 开始调优。")

import tensorflow as tf
from tensorflow.keras import metrics

class CustomF1Score(metrics.Metric):
    def __init__(self, name='custom_f1_score', threshold=0.5, **kwargs):
        super().__init__(name=name, **kwargs)
        self.threshold = threshold
        # 初始化用于计算指标的状态变量
        self.true_positives = self.add_weight(name='tp', initializer='zeros')
        self.false_positives = self.add_weight(name='fp', initializer='zeros')
        self.false_negatives = self.add_weight(name='fn', initializer='zeros')

    def update_state(self, y_true, y_pred, sample_weight=None):
        # 将预测值和真实值转换为布尔类型，并根据阈值进行二值化
        y_true = tf.cast(y_true, tf.bool)
        y_pred = y_pred > self.threshold

        # 计算TP, FP, FN
        tp = tf.cast(tf.reduce_sum(tf.cast(y_true & y_pred, tf.float32)), tf.float32)
        fp = tf.cast(tf.reduce_sum(tf.cast(~y_true & y_pred, tf.float32)), tf.float32)
        fn = tf.cast(tf.reduce_sum(tf.cast(y_true & ~y_pred, tf.float32)), tf.float32)

        # 更新状态变量
        self.true_positives.assign_add(tp)
        self.false_positives.assign_add(fp)
        self.false_negatives.assign_add(fn)

    def result(self):
        # 根据累积的TP, FP, FN计算F1 Score
        precision = self.true_positives / (self.true_positives + self.false_positives + tf.keras.backend.epsilon())
        recall = self.true_positives / (self.true_positives + self.false_negatives + tf.keras.backend.epsilon())
        f1 = 2 * (precision * recall) / (precision + recall + tf.keras.backend.epsilon())
        return f1

    def reset_state(self):
        # 重置状态变量，为下一个epoch或评估周期做准备
        self.true_positives.assign(0.0)
        self.false_positives.assign(0.0)
        self.false_negatives.assign(0.0)

# 在MyHyperModel中使用自定义指标
# class MyHyperModel(...):
#     def build(self, hp):
#         # ... (模型定义)
#         model.compile(
#             # ...
#             metrics=[CustomF1Score(name='my_custom_f1')] # 使用自定义指标，并指定日志名称
#         )
#         return model
#
# tuner = kt.RandomSearch(
#     MyHyperModel(),
#     objective=kt.Objective("val_my_custom_f1", direction="max"), # KerasTuner目标名称
#     # ...
# )