1D Convolutional Neural Network Input Shape `ValueError`

Question

I am generating a 1D Convolution and have some trouble with the input shape of my data. I had a look at some posts and it seems the error was that the data has to be 3D but my data is already 3D.

# shape
# x_train shape: (1228, 1452, 20)
# y_train shape: (1228, 1452, 8)
# x_val shape: (223, 680, 20)
# x_val shape: (223, 680, 8)

###
n_outputs = 8
n_timesteps = 1452
n_features = 20

model = Sequential()
model.add(Conv1D(filters=64, kernel_size=3, activation='relu', input_shape=(x_train.shape[1:]))) # ie 1452, 20
model.add(Conv1D(filters=64, kernel_size=3, activation='relu'))
model.add(Dropout(0.5))
model.add(MaxPooling1D(pool_size=2))
model.add(Flatten())
model.add(Dense(100, activation='relu'))
model.add(Dense(n_outputs, activation='softmax'))
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])

model.fit(x_train, y_train, epochs=9, 
                batch_size=64, 
                shuffle=True)

But I keep on getting this error message:

ValueError: A target array with shape (1228, 1452, 8) was passed for an output of shape (None, 8) while using as loss `categorical_crossentropy`. This loss expects targets to have the same shape as the output.

What I gather from this is that target shape which is 3 dimensional is not the same as the 2 dimensional output and so it can't work out the loss, but I just need to find a way to reshape so that they are equal.

EDIT model.summary() is shown below

_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv1d (Conv1D)              (None, 1450, 64)          3904      
_________________________________________________________________
conv1d_1 (Conv1D)            (None, 1448, 64)          12352     
_________________________________________________________________
dropout (Dropout)            (None, 1448, 64)          0         
_________________________________________________________________
max_pooling1d (MaxPooling1D) (None, 724, 64)           0         
_________________________________________________________________
flatten (Flatten)            (None, 46336)             0         
_________________________________________________________________
dense (Dense)                (None, 100)               4633700   
_________________________________________________________________
dense_1 (Dense)              (None, 8)                 808       
=================================================================
Total params: 4,650,764
Traceback (most recent call last):
Trainable params: 4,650,764
Non-trainable params: 0

The only thing that jumps out at me is possibly your input_shape may need another comma to make the batch size ambiguous. Perhaps input_shape=(1452,20,) <-- notice the extra comma — TheLoneDeranger
– TheLoneDeranger, Commented May 17, 2019 at 16:41
:( That's because I was being oblivious. It's just like the error says: your y values have an unexpected dimension. Notice that y_train shape: 1228,1452,8 but down at your last Dense layer you've got n_outputs=8. What's actually hitting your output is (batch,1452,8) but it's expecting (batch,8) — TheLoneDeranger
– TheLoneDeranger, Commented May 17, 2019 at 16:58
:) Precisely that, then: your y values should be (only) your one-hot encoded vectors of length 8. The network will only welcome inputs of shape (batch,1452,20) and only outputs shape (batch,8). y_train is (batch,1452,8) but needs to be (batch,8). Also, your training and validation shapes don't agree: (batch,1452,20) for training and (batch,680,20) for validation, but (1452 != 680). — TheLoneDeranger
– TheLoneDeranger, Commented May 17, 2019 at 17:47

Haj Sai · Accepted Answer · 2019-05-17 22:58:34Z

1

The issue in my case was that the target vector was 3D whilst the output vector was 2D and so there is an obvious mismatch. To fix the issue change the shape of the y_train to (batch, 8) or use return_sequences=True to return the same shape from the previous LSTM layer.

answered May 17, 2019 at 22:58

Haj Sai

3111 gold badge4 silver badges13 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

1D Convolutional Neural Network Input Shape `ValueError`

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related