KerasTuner中自定义指标（如F1、AUC）作为调优目标的实践指南

import tensorflow as tf
from tensorflow.keras import metrics
# ... 其他导入 ...

class MyHyperModel(kt.HyperModel):
    def build(self, hp):
        model = Sequential()
        # ... 模型层定义 ...
        model.add(layers.Dense(1, activation="sigmoid"))

        # 编译模型时包含F1Score
        model.compile(
            optimizer=Adam(learning_rate=hp.Float('learning_rate', 5e-5, 5e-1, step=0.001)),
            loss='binary_crossentropy',
            metrics=['accuracy', metrics.F1Score(thresholds=0.5)] # 添加F1Score实例
        )
        return model

tuner = kt.RandomSearch(
    MyHyperModel(),
    objective=kt.Objective("val_f1_score", direction="max"), # 使用val_f1_score作为目标
    max_trials=100,
    overwrite=True,
    directory="my_dir",
    project_name="tune_hypermodel",
)

import keras_tuner as kt
import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.metrics import classification_report
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
from tensorflow.keras.regularizers import l2
from tensorflow.keras import metrics # 导入metrics模块

# 模拟数据
# 为了使示例可运行，我们创建一些虚拟数据
num_samples = 1000
num_features = 10
X = pd.DataFrame(tf.random.normal((num_samples, num_features)).numpy())
y = pd.DataFrame(tf.random.uniform((num_samples, 1), minval=0, maxval=2, dtype=tf.int32)).iloc[:, 0]
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

class MyHyperModel(kt.HyperModel):
    def build(self, hp):
        model = Sequential()
        model.add(layers.Flatten(input_shape=(X_train.shape[1],))) # 确保输入形状正确
        model.add(
            layers.Dense(
                units=hp.Int("units", min_value=24, max_value=128, step=10),
                activation="relu",
                kernel_regularizer=l2(hp.Float('l2_reg', 1e-5, 1e-2, sampling='log')) # 添加L2正则化作为超参数
            )
        )
        model.add(layers.Dense(1, activation="sigmoid"))  # 二分类输出层

        # 编译模型，包含F1Score作为评估指标
        model.compile(
            optimizer=Adam(learning_rate=hp.Float('learning_rate', 5e-5, 5e-1, sampling='log')),
            loss='binary_crossentropy',
            metrics=['accuracy', metrics.F1Score(thresholds=0.5)] # 添加F1Score
        )
        return model

    def fit(self, hp, model, *args, **kwargs):
        return model.fit(
            *args,
            batch_size=hp.Choice("batch_size", [16, 32, 64]), # 调整batch_size选项
            epochs=hp.Int('epochs', min_value=5, max_value=25, step=5),
            **kwargs,
        )

# 初始化KerasTuner的RandomSearch
tuner = kt.RandomSearch(
    MyHyperModel(),
    objective=kt.Objective("val_f1_score", direction="max"), # 目标设置为val_f1_score，方向为最大化
    max_trials=10, # 减少试验次数以便快速运行示例
    overwrite=True,
    directory="my_dir",
    project_name="tune_hypermodel_f1",
)

# 开始搜索
print("开始搜索超参数...")
tuner.search(X_train, y_train, validation_data=(X_test, y_test), callbacks=[keras.callbacks.EarlyStopping('val_loss', patience=3)])
print("搜索完成。")

# 获取最佳超参数和模型
best_hps = tuner.get_best_hyperparameters(num_trials=1)[0]
best_model = tuner.get_best_models(num_models=1)[0]

print(f"\n最佳超参数: {best_hps.values}")
loss, accuracy, f1_score = best_model.evaluate(X_test, y_test, verbose=0)
print(f"最佳模型在测试集上的F1 Score: {f1_score:.4f}")