Python - Keras：LSTM 数据结构 ValueError

Question

我正在尝试使用 return_sequence 训练 LSTM 模型以返回每个输入时间步的隐藏状态输出，从而解决回归问题。

我的数据形状是：(31, 2720, 16)，即 31 批 2720 个样本，具有 16 个特征。
我的目标形状是： (31, 2720, 1) 即 31 批 2720 行包含 1 个值。

我已经建立了以下模型：

model = Sequential()
opt = Adam(learning_rate=0.0001, clipnorm=1)

num_samples = train_x.shape[1]
num_features = train_x.shape[2]

model.add(Masking(mask_value=-10., input_shape=(num_samples, num_features)))


model.add(LSTM(32, return_sequences=True, stateful=False, activation='tanh'))
model.add(Dropout(0.3))

#this is the last LSTM layer, use return_sequences=False
model.add(LSTM(16, return_sequences=False, stateful=False,  activation='tanh'))
model.add(Dropout(0.3))
model.add(Dense(16, activation='tanh'))
model.add(Dense(8, activation='tanh'))
model.add(Dense(1, activation='sigmoid'))

model.compile(loss='mse', optimizer='adam' ,metrics=[metrics.mean_absolute_error, metrics.mean_squared_error])

logdir = os.path.join(logs_base_dir, datetime.datetime.now().strftime("%Y%m%d-%H%M%S"))
tensorboard_callback = TensorBoard(log_dir=logdir, update_freq=1)
model.summary()

总结：

Model: "sequential_33"

_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
masking_24 (Masking)         (None, 2720, 16)          0         
_________________________________________________________________
lstm_61 (LSTM)               (None, 2720, 32)          6272      
_________________________________________________________________
dropout_51 (Dropout)         (None, 2720, 32)          0         
_________________________________________________________________
lstm_62 (LSTM)               (None, 16)                3136      
_________________________________________________________________
dropout_52 (Dropout)         (None, 16)                0         
_________________________________________________________________
dense_67 (Dense)             (None, 16)                272       
_________________________________________________________________
dense_68 (Dense)             (None, 8)                 136       
_________________________________________________________________
dense_69 (Dense)             (None, 1)                 9         
=================================================================
Total params: 9,825
Trainable params: 9,825
Non-trainable params: 0
_________________________________________________________________

尝试拟合模型时，出现以下错误：

---

ValueError                                Traceback (most recent call last)
<ipython-input-354-afdba8dea179> in <module>()
----> 1 model.fit(train_x, train_y, epochs=1000, batch_size=128,validation_split = 0.2, callbacks=[tensorboard_callback,checkpoint])

5 frames
/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/keras/engine/training_utils.py in check_loss_and_target_compatibility(targets, loss_fns, output_shapes)
    808           raise ValueError('A target array with shape ' + str(y.shape) +
    809                            ' was passed for an output of shape ' + str(shape) +
--> 810                            ' while using as loss `' + loss_name + '`. '
    811                            'This loss expects targets to have the same shape '
    812                            'as the output.')

我正在尝试掌握构建数据的正确方法，我缺少什么？

Answer 1

您的目标形状为(31, 2720, 1)，您当前模型的输出形状为(31, 1)。这种情况下的错误是自我解释的。

您可以通过以下两种方式之一解决此问题：

1. 看看你的模型，我猜你只想要关于最后一个序列的损失。这种情况下，可以拨打model.fit如下：

   model.fit(train_x, train_y[:, -1, :], ...) 
   

2. 如果要计算所有时间步的损失，请将return_sequences=True 添加到第二个 LSTM 层：

   model.add(LSTM(16, return_sequences=True, stateful=False,  activation='tanh'))
   

Answer 2

您想要的目标形状与模型输出不一致。

改变这一行

model.add(LSTM(16, return_sequences=False, stateful=False, activation='tanh'))

到

model.add(LSTM(16, return_sequences=True, stateful=False, activation='tanh'))

所以时间维度就在那里。

from tensorflow.keras.layers import *
from tensorflow.keras.models import Model, Sequential
from tensorflow.keras.optimizers import Adam
import tensorflow as tf
import numpy as np

from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense

from tensorflow.keras import metrics
import tensorflow as tf
import numpy as np 

model = Sequential()
opt = Adam(learning_rate=0.0001, clipnorm=1)

num_samples = 2720
num_features = 16

model.add(Masking(mask_value=-10., input_shape=(num_samples, num_features)))


model.add(LSTM(32, return_sequences=True, stateful=False, activation='tanh'))
model.add(Dropout(0.3))

#this is the last LSTM layer, use return_sequences=False
model.add(LSTM(16, return_sequences=True, stateful=False,  activation='tanh'))
model.add(Dropout(0.3))
model.add(Dense(16, activation='tanh'))
model.add(Dense(8, activation='tanh'))
model.add(Dense(1, activation='sigmoid'))

model.compile(loss='mse', optimizer='adam' ,metrics=[metrics.mean_absolute_error, metrics.mean_squared_error])

model.summary()

Model: "sequential_3"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
masking_1 (Masking)          (None, 2720, 16)          0         
_________________________________________________________________
lstm_2 (LSTM)                (None, 2720, 32)          6272      
_________________________________________________________________
dropout_2 (Dropout)          (None, 2720, 32)          0         
_________________________________________________________________
lstm_3 (LSTM)                (None, 2720, 16)          3136      
_________________________________________________________________
dropout_3 (Dropout)          (None, 2720, 16)          0         
_________________________________________________________________
dense_3 (Dense)              (None, 2720, 16)          272       
_________________________________________________________________
dense_4 (Dense)              (None, 2720, 8)           136       
_________________________________________________________________
dense_5 (Dense)              (None, 2720, 1)           9         
=================================================================
Total params: 9,825
Trainable params: 9,825
Non-trainable params: 0