[英]Keras Conv1d Input Shape: Error when checking input
我正在使用帶有 TF 后端的 keras 來構建一個簡單的Conv1d
網絡。 數據具有以下形狀:
train feature shape: (33960, 3053, 1)
train label shape: (33960, 686, 1)
我用以下方法構建模型:
def create_conv_model():
inp = Input(shape=(3053, 1))
conv = Conv1D(filters=2, kernel_size=2)(inp)
pool = MaxPool1D(pool_size=2)(conv)
flat = Flatten()(pool)
dense = Dense(686)(flat)
model = Model(inp, dense)
model.compile(loss='mse', optimizer='adam')
return model
型號概要:
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_1 (InputLayer) (None, 3053, 1) 0
_________________________________________________________________
conv1d_1 (Conv1D) (None, 3052, 2) 6
_________________________________________________________________
max_pooling1d_1 (MaxPooling1 (None, 1526, 2) 0
_________________________________________________________________
flatten_1 (Flatten) (None, 3052) 0
_________________________________________________________________
dense_1 (Dense) (None, 686) 2094358
=================================================================
Total params: 2,094,364
Trainable params: 2,094,364
Non-trainable params: 0
運行時
model.fit(x=train_feature,
y=train_label_categorical,
epochs=100,
batch_size=64,
validation_split=0.2,
validation_data=(test_feature,test_label_categorical),
callbacks=[tensorboard,reduce_lr,early_stopping])
我收到以下非常常見的錯誤:
ValueError: Error when checking input: expected input_1 to have 3 dimensions, but got array with shape (8491, 3053)
我已經檢查了幾乎所有關於這個非常常見問題的帖子,但我一直無法找到解決方案。 我究竟做錯了什么? 我不明白發生了什么。 形狀(8491, 3053)
來自哪里?
任何幫助將不勝感激,我無法讓它消失。
將model.fit
函數中的validation_data=(test_feature,test_label_categorical)
model.fit
為
validation_data=(np.expand_dims(test_feature, -1),test_label_categorical)
該模型需要形狀(8491, 3053, 1)
驗證功能,但在上面的代碼中,您提供的是(8491, 3053)
。
聲明:本站的技術帖子網頁,遵循CC BY-SA 4.0協議,如果您需要轉載,請注明本站網址或者原文地址。任何問題請咨詢:yoyou2525@163.com.