使用CNN超参数优化进行多目标回归时出错

Question

我正在尝试使用CNN解决多目标回归问题。 为了优化超参数，我使用了Talos库，但最终出现一个错误：

检查输入时出错：预期conv1d_1_input具有3维，但数组的形状为（280000，70）

数据集描述：X具有70个特征，Y具有32个目标时间序列长度：509760个时间点

Shape of Dataset:

Train Data size: (400000, 70),
Train Target size: (400000, 32),
Test Data size: (109760, 70),
Test Target size: (109760, 32),

码：

tf.keras.backend.clear_session()

def model(x_train, y_train, x_val, y_val, params):

    n_timesteps, n_features, n_outputs = x_train.shape[0],    x_train.shape[1], y_train.shape[1]

    model = Sequential()

    model.add(Conv1D(filters=params['conv1_filter'], kernel_size=(3), activation=params['activation'], input_shape=(n_timesteps,n_features)))
    model.add(Conv1D(filters=params['conv2_filter'], kernel_size=(3), activation=params['activation']))
    model.add(MaxPooling1D(pool_size=2))
    model.add(Dropout(params['dropout']))
    model.add(Conv1D(filters=8, kernel_size=3, activation=params['activation']))
    model.add(MaxPooling1D(pool_size=2))
    model.add(Dense(100, activation=params['activation']))
    model.add(Dense(n_outputs))

    model.add(Flatten())    
    model.compile(loss='mse',optimizer=params['optimizer'](),metrics=['acc', fmeasure_acc])


    out = model.fit(x_train, y_train, validation_data=[x_val, y_val], batch_size=params['batch_size'], epochs=params['epochs'],verbose=1,shuffle=True)

    return out, model

p = {'conv1_filter':[32],
     'conv2_filter':[32],
     'optimizer': [Nadam],
     'batch_size': [1000],
     'epochs': [5],
     'dropout': [0.5],
     'activation':[relu]}

# and run the experiment
t = ta.Scan(x=x_train,y=y_train,model=model,params=p,experiment_no='1')

错误如下：

  168 # input parameters section ends 169 --> 170 self._null = self.runtime() 171 172 def runtime(self): /anaconda3/lib/python3.7/site-packages/talos/scan/Scan.py in runtime(self) 173 174 self = scan_prepare(self) --> 175 self = scan_run(self) /anaconda3/lib/python3.7/site-packages/talos/scan/scan_run.py in scan_run(self) 16 disable=self.disable_progress_bar) 17 while len(self.param_log) != 0: ---> 18 self = scan_round(self) 19 self.pbar.update(1) 20 self.pbar.close() /anaconda3/lib/python3.7/site-packages/talos/scan/scan_round.py in scan_round(self) 30 # fit the model 31 try: ---> 32 _hr_out, self.keras_model = ingest_model(self) 33 except TypeError as err: 34 if err.args[0] == "unsupported operand type(s) for +: 'int' and 'numpy.str_'": /anaconda3/lib/python3.7/site-packages/talos/model/ingest_model.py in ingest_model(self) 8 self.x_val, 9 self.y_val, ---> 10 self.round_params) <ipython-input-9-b33b869928ed> in model(x_train, y_train, x_val, y_val, params) 20 21 ---> 22 out = model.fit(x_train, y_train, validation_data=[x_val, y_val], batch_size=params['batch_size'], epochs=params['epochs'],verbose=1,shuffle=True) 23 24 /anaconda3/lib/python3.7/site-packages/keras/engine/training.py in fit(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, **kwargs) 950 sample_weight=sample_weight, 951 class_weight=class_weight, --> 952 batch_size=batch_size) 953 # Prepare validation data. 954 do_validation = False /anaconda3/lib/python3.7/site-packages/keras/engine/training.py in _standardize_user_data(self, x, y, sample_weight, class_weight, check_array_lengths, batch_size) 749 feed_input_shapes, 750 check_batch_axis=False, # Don't enforce the batch size. --> 751 exception_prefix='input') 752 753 if y is not None: /anaconda3/lib/python3.7/site-packages/keras/engine/training_utils.py in standardize_input_data(data, names, shapes, check_batch_axis, exception_prefix) 126 ': expected ' + names[i] + ' to have ' + 127 str(len(shape)) + ' dimensions, but got array ' --> 128 'with shape ' + str(data_shape)) 129 if not check_batch_axis: 130 data_shape = data_shape[1:] ValueError: Error when checking input: expected conv1d_1_input to have 3 dimensions, but got array with shape (280000, 70) 

Answer 1

我再次查看了一下，需要在代码中进行一些更改。 我测试了它，现在应该可以工作了：

首先，模型本身存在两个问题。

即，模型的输入应为(n_features, 1)而不是(n_timesteps,n_features) 。 Keras内部认为，给定输入形状时，您会忽略批次尺寸。 因此，何时将其注册为形状是(batch_size, n_features, 1) 。 最后一个尺寸是必需的，以便与Conv1D兼容。

它应该是这样的：

model.add(Conv1D(filters=params['conv1_filter'], kernel_size=(3), activation=params['activation'], input_shape=(n_features, 1))) 

其次，模型不应以Flatten()层结尾。 相反，该层应位于Dense()层之前（以便它看到2D输入而不是3D输入）。 两次更改后，模型应如下所示：

model = Sequential()

model.add(Conv1D(filters=params['conv1_filter'], kernel_size=(3), activation=params['activation'], input_shape=(n_features,1))) 
model.add(Conv1D(filters=params['conv2_filter'], kernel_size=(3), activation=params['activation']))
model.add(MaxPooling1D(pool_size=2))
model.add(Dropout(params['dropout']))
model.add(Conv1D(filters=8, kernel_size=3, activation=params['activation']))
model.add(MaxPooling1D(pool_size=2))
model.add(Flatten())
model.add(Dense(100, activation=params['activation']))
model.add(Dense(n_outputs))

由于我们更改了模型的input_shape ，因此需要更改数据的形状。 有什么需要做的是改变(400000, 70)至(400000, 70, 1)和(109760, 70)至(109760, 70, 1) 可以使用以下命令完成此操作：

x_train = np.expand_dims(x_train, axis=-1)
x_val = np.expand_dims(x_val, axis=-1)

我没有在Talos上测试它，但是keras部分工作得很好。