检查目标时出错：预期 dense_24 有 3 个维度，但得到了形状为 (3283, 1) 的数组

Question

我有一个 X_train = [(4096, 18464),(4097, 43045),(4098, 38948),(4099, 2095),(4100, 59432),(4101, 55338),(4102, 51245) ,(4103, 26658),(4104, 30755),....] 形状为 (3283, 2) 和

y_train = [19189, 19189, 19189, ..., 1155085434105692417, 1155120620365152513,...] 形状为 (3283, 1)

我使用代码重塑了 X_train：

X_train = np.reshape(X_train, (X_train.shape[0], 1, X_train.shape[1]))
X_test = np.reshape(X_test, (X_test.shape[0], 1, X_test.shape[1]))

得到形状 (3283, 1, 2)

现在我建立了一个 lstm 模型：

data_dim= 2
timesteps=1
num_classes=2

model_pass = Sequential()
model_pass.add(LSTM(units=64,  return_sequences=True, 
                input_shape=(timesteps, data_dim)))
model_pass.add(Dense(2, activation='sigmoid'))
model_pass.compile(loss='binary_crossentropy', optimizer='adam',metrics=['accuracy'])
model_pass.summary()
model_pass.fit(X_train, y_train,batch_size=1, epochs = 1, verbose = 1)

但这给了我一个错误： ValueError: 检查目标时出错：预期dense_24 有3 维，但得到的数组形状为(3283, 1)

谁能告诉我该怎么办？

Answer 1

在密集层之后，输出形状为(number of samples, timesteps, 2)。数字 2 来自Dense(2,...)。但 y_train_pass 的形状可能为 (number of samples, 1)。这给出了一个错误。

这是我将 Dense(2,...) 更改为 Dense(1,...) 并重塑 y_train 的可能代码示例

from keras.models import Sequential
from keras.layers import LSTM
from keras.layers import Dense
import numpy as np

X_train = np.random.randn(3283,2)
X_test = np.random.randn(1000,2)
y_train = np.random.randint(2, size=(3283,1))
print(y_train.shape)
X_train = np.reshape(X_train, (X_train.shape[0], 1, X_train.shape[1]))
X_test = np.reshape(X_test, (X_test.shape[0], 1, X_test.shape[1]))
y_train = np.reshape(y_train, (y_train.shape[0],1, y_train.shape[1]))

data_dim= 2
timesteps=1
num_classes=2

model_pass = Sequential()
model_pass.add(LSTM(units=64,  return_sequences=True, 
                input_shape=(timesteps, data_dim)))
model_pass.add(Dense(1, activation='sigmoid'))
model_pass.compile(loss='binary_crossentropy', optimizer='adam',metrics=['accuracy'])
model_pass.summary()
model_pass.fit(X_train, y_train,batch_size=1, epochs = 1, verbose = 1)

顺便说一句，只是一个评论。奇怪的是，您的 y_train 值没有“二进制”值，而您使用的是 loss='binary_crossentropy'。