ValueError: Error when checking target: expected dense_2 to have 4 dimensions, but got array with shape (7942, 1)

Question

I have been using the following functional API for an image classification task using CNN:

def create_model(X_train, X_test):

    visible = Input(shape=(X_train.shape[0], X_train.shape[1], 1))
    conv1 = Conv2D(32, kernel_size=4, activation='relu')(visible)
    hidden1 = Dense(10, activation='relu')(pool2)
    output = Dense(1, activation='sigmoid')(hidden1)

    model = Model(inputs = visible, outputs = output)

    model.compile(loss='binary_crossentropy',
              optimizer='rmsprop',
              metrics=['accuracy'])

    return model

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

model = create_model(X_train, X_test)

model.fit(X_tr, y_train, validation_split = 0.1, batch_size=10, epochs=10, verbose = 1, callbacks=[EarlyStopping(patience=5,verbose=1)]) 

where, X_train is a 7942*6400 dimensional list and y_train being a 1-D list with corresponding 7942 labels.

The error:

ValueError: Error when checking target: expected dense_2 to have 4 dimensions, but got array with shape (7942, 1)

What has possibly gone wrong here as I am a newbie to the functional API?

Answer 1

The message says that y_train is not compatible with the model's output.

Your model is outputting (None, width, height, 1). You should add a Flatten() layer after the convolution to make the data have only 2 dimensions from this point on.

---

Additional comments:

The input data must have a shape compatible with the model.

The shape of X_train must be (7942,80,80,1)
The input_shape of the model must be (80,80,1)

If you use a (1,6400, 1) shape, your Conv2D layer will be pretty useless, because it will not be able to intepret the data as a 2D image.