使用KerasRegressor时遇AttributeError报错求助（TF2.15/Keras2.15）

我正在做学习练习，尝试用KerasRegressor为简单序列神经网络优化超参数，代码如下：

from sklearn.model_selection import GridSearchCV, RandomizedSearchCV
from scipy.stats import randint as sp_randint
from keras.wrappers.scikit_learn import KerasRegressor
from sklearn.metrics import mean_squared_error, make_scorer

'''CREATE THE MODEL'''
def design_model(features):
    model = Sequential(name = "My_Sequential_Model") 
    model.add(InputLayer(input_shape=(features.shape[1],))) 
    model.add(Dense(128, activation='relu')) 
    model.add(Dense(1)) 
    opt = Adam(learning_rate=0.01)
    model.compile(loss='mse', metrics=['mae'], optimizer=opt) 
    return model

'''TEST/PLOT THE MODEL: GRID SEARCH'''
def do_grid_search():
    batch_size = [6, 64]
    epochs = [10, 30, 61]
    model = KerasRegressor(build_fn=design_model, features=features_train) # KerasRegressor expects a function and not the model
    param_grid = dict(batch_size=batch_size, epochs=epochs)
    grid = GridSearchCV(estimator = model, param_grid=param_grid, scoring = make_scorer(mean_squared_error, greater_is_better=False),return_train_score = True)
    grid_result = grid.fit(features_train, labels_train, verbose = 0)
    print(grid_result)
    print("Best: %f using %s" % (grid_result.best_score_, grid_result.best_params_))

    means = grid_result.cv_results_['mean_test_score']
    stds = grid_result.cv_results_['std_test_score']
    params = grid_result.cv_results_['params']
    for mean, stdev, param in zip(means, stds, params): print("%f (%f) with: %r" % (mean, stdev, param))

    print("Traininig")
    means = grid_result.cv_results_['mean_train_score']
    stds = grid_result.cv_results_['std_train_score']
    for mean, stdev, param in zip(means, stds, params): print("%f (%f) with: %r" % (mean, stdev, param))

print("-------------- GRID SEARCH --------------------")
do_grid_search()

运行后持续触发以下错误：

Traceback (most recent call last):
  File "C:\Users\User\AppData\Local\Programs\Python\Python310\lib\site-packages\sklearn\model_selection\_validation.py", line 732, in _fit_and_score
    estimator.fit(X_train, y_train, **fit_params)
  File "C:\Users\User\AppData\Local\Programs\Python\Python310\lib\site-packages\keras\wrappers\scikit_learn.py", line 167, in fit
AttributeError: module 'keras.losses' has no attribute 'is_categorical_crossentropy'

使用环境：Tensorflow 2.15、Keras 2.15版本

1. 替换KerasRegressor的导入来源 #

错误核心是keras.wrappers.scikit_learn下的旧版包装器调用了Keras 2.15已移除的APIkeras.losses.is_categorical_crossentropy，改用TensorFlow官方维护的包装器即可解决：

# 替换原来的导入
from tensorflow.keras.wrappers.scikit_learn import KerasRegressor

2. 补充缺失的Keras组件导入 #

原代码未导入核心模型组件，需补充：

from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import InputLayer, Dense
from tensorflow.keras.optimizers import Adam

3. 修正模型构建函数的参数传递 #

原代码直接传入训练特征的方式不符合KerasRegressor规范，改为传递输入维度参数：

def design_model(input_dim):
    model = Sequential(name="My_Sequential_Model") 
    model.add(InputLayer(input_shape=(input_dim,))) 
    model.add(Dense(128, activation='relu')) 
    model.add(Dense(1)) 
    opt = Adam(learning_rate=0.01)
    model.compile(loss='mse', metrics=['mae'], optimizer=opt) 
    return model

4. 调整KerasRegressor的初始化方式 #

创建KerasRegressor时，传递输入维度而非原始训练数据：

model = KerasRegressor(build_fn=design_model, input_dim=features_train.shape[1])

from sklearn.model_selection import GridSearchCV, RandomizedSearchCV
from scipy.stats import randint as sp_randint
from tensorflow.keras.wrappers.scikit_learn import KerasRegressor
from sklearn.metrics import mean_squared_error, make_scorer
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import InputLayer, Dense
from tensorflow.keras.optimizers import Adam

'''CREATE THE MODEL'''
def design_model(input_dim):
    model = Sequential(name="My_Sequential_Model") 
    model.add(InputLayer(input_shape=(input_dim,))) 
    model.add(Dense(128, activation='relu')) 
    model.add(Dense(1)) 
    opt = Adam(learning_rate=0.01)
    model.compile(loss='mse', metrics=['mae'], optimizer=opt) 
    return model

'''TEST/PLOT THE MODEL: GRID SEARCH'''
def do_grid_search():
    batch_size = [6, 64]
    epochs = [10, 30, 61]
    model = KerasRegressor(build_fn=design_model, input_dim=features_train.shape[1])
    param_grid = dict(batch_size=batch_size, epochs=epochs)
    # 可简化scoring为sklearn内置的neg_mean_squared_error
    grid = GridSearchCV(estimator=model, param_grid=param_grid, 
                        scoring='neg_mean_squared_error', return_train_score=True)
    grid_result = grid.fit(features_train, labels_train, verbose=0)
    print(grid_result)
    print("Best: %f using %s" % (grid_result.best_score_, grid_result.best_params_))

    means = grid_result.cv_results_['mean_test_score']
    stds = grid_result.cv_results_['std_test_score']
    params = grid_result.cv_results_['params']
    for mean, stdev, param in zip(means, stds, params): 
        print("%f (%f) with: %r" % (mean, stdev, param))

    print("Training")
    means = grid_result.cv_results_['mean_train_score']
    stds = grid_result.cv_results_['std_train_score']
    for mean, stdev, param in zip(means, stds, params): 
        print("%f (%f) with: %r" % (mean, stdev, param))

print("-------------- GRID SEARCH --------------------")
do_grid_search()

内容的提问来源于stack exchange，提问作者Code Monkey