CODEWITHSUNDEEP
pythonmachine-learningkerasneural-networkconv-neural-network
Mr. Nobody
Mr. Nobody

Reputation: 240

ValueError: Error when checking target: expected dense_16 to have shape (1,) but got array with shape (30,)

I'm having this error, and from all I've seen from answers to similar questions it is due to not specifying the last layer to have 30 neurons, since your desired output is an array with size 30. But this is not my case, I've specified this, so I find it very weird that this error pops up if the number of neurons and output shape match perfectly. Below there's the part of the code that may be causing this error (images_gray is a numpy array consisting of greyscale images with shape (2000, 128, 128) and result_array is a list of numpy arrays of size 30).

images_gray = np.reshape(images_gray, (len(images_gray), img_rows, img_cols, 1))
to_shuffle = list(zip(images_gray, result_array))
random.shuffle(to_shuffle)
im_input, res_output = zip(*to_shuffle)
im_input, res_output = np.array(im_input[:10]), np.array(res_output[:10])
img_rows, img_cols, _ = np.shape(im_input[0])
input_shape = (img_rows, img_cols, 1)
if K.image_data_format() == 'channels_first':
    input_shape = (1, img_rows, img_cols)
model = buildCNN(input_shape)
adadelta = keras.optimizers.Adadelta(lr=0.5)
model.compile(loss="sparse_categorical_crossentropy",optimizer=adadelta,metrics=['accuracy'])
model.fit(im_input, res_output, batch_size=128, epochs=10, verbose=1)

Here's the buildCNN() function:

def buildCNN(input_shape):
    model = Sequential()
    model.add(Conv2D(32, kernel_size=(3,3),activation='relu',input_shape=input_shape))
    model.add(Conv2D(64, (3,3), activation='relu'))
    model.add(MaxPooling2D(pool_size=(2,2)))
    model.add(Dropout(0.25))
    model.add(Flatten())
    model.add(Dense(128,activation='relu'))
    model.add(Dropout(0.5))
    model.add(Dense(30, activation='sigmoid'))
    return model

And what I get when I use model.sumary():

enter image description here

Upvotes: 0

Views: 267

Answers (1)

Matus Dubrava
Matus Dubrava

Reputation: 14462

If your target (y values) have length of 30 then I assume that you want to do multiclass classification. In that case, your are using wrong activation function in your output layer which should be softmax instead of sigmoid.

model.add(Dense(30, activation='softmax'))

sigmoid is used with binary classification.

And if you want to do binary classification, then your target (y values) should have a length of 1, instead of 30.

Also note that you will most likely receive another error because if your target values are one-hot encoded (assuming from the fact that they are arrays of the length of 30) then the correct loss function to pick is categorical_crossentropy.

sparse_categorical_crossentropy loss is used for multiclass classification when labels are provided as integers (not one-hot encoded).

So in the case of one-hot encoded labels, the compile method should be called like this

model.compile(loss="categorical_crossentropy",optimizer=adadelta,metrics=['accuracy'])

Upvotes: 1

Related Questions