神经网络的网格搜索超参数Keras 错误代码

编程问答 2022-06-11

问题描述

我正在尝试同时使用Keras和SKlearn优化神经网络的超参数，我正在用KerasRegressor包装我的模型，因为这是一个回归问题。我正在尝试优化各种参数：

  1. 隐藏层数
  2. 每个隐藏层的神经元数量
  3. 优化程序
  4. 纪元数和Batch_Size
  5. 激活功能

我可以单独执行此操作，但我想尝试同时运行所有内容和所有参数:)来查看是否有可能进行全面比较，但我不断遇到以下错误。

错误

PicklingError: Could not pickle the task to send it to the workers.

在这里您可以在下面找到我的代码

def build_regressor(optimizer='adam',activation = 'relu',hidden_layers=1,neurons=1):
  # Initialize the constructor
    regressor = Sequential()
  # Add an input layer
    regressor.add(Dense(neurons,activation=activation,input_shape = (x_train.shape[1],)))

    for i in range(hidden_layers):
      # Add one hidden layer
        regressor.add(Dense(neurons,activation=activation))

  # Add an output layer 
    regressor.add(Dense(1,activation=activation))
  #compile model
    regressor.compile(loss='mean_squared_error',optimizer= optimizer,metrics= ['mse','mae'],epochs = epochs,batch_size = batch_size)
    return model

#Wrap Model
regressor = KerasRegressor(build_fn=build_regressor,verbose=1)

# define the grid search parameters
batch_size = [10,20,40,60,80,100,150]
epochs = [10,50,150,200]
neurons = [1,32,64,128,256,512]
optimizer = ['SGD','adam']
#activation = ['relu','linear','softmax']

param_grid = dict(batch_size = batch_size,neurons = neurons,optimizer = optimizer)

#implement grid_search
grid = GridSearchCV(estimator=regressor,param_grid=param_grid,n_jobs=-1,cv=3,scoring = 'r2')
grid_result = grid.fit(x_train,y_train)

#Fit Model
results=regressor.fit(x_train,y_train)
y_pred= regressor.predict(x_valid)

# summarize results
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,param))

代码

解决方法

暂无找到可以解决该程序问题的有效方法，小编努力寻找整理中！

如果你已经找到好的解决方法，欢迎将解决方案带上本链接一起发送给小编。

小编邮箱:dio#foxmail.com (将#修改为@）

hyperparameterskerasmachine-learningpythontensorflow