CODEWITHSUNDEEP
pythonmachine-learningkerasconv-neural-network
RinW
RinW

Reputation: 553

Incompatible input data error in keras, dimensions mismatch ValueError

I'm trying to understand how to write a Conv1D model on Keras but I keep running into a dimensionality mismatch error. Right now x_train = [1000,294] (1000 items and 294 features) and y_train = [1000,9] (1000 items and 9 labels). I keep getting the error message expected 3 dimensions but got 2 dimensions. But I try to fix it, it appears again. Some issues on github has suggested Flatten() but it is already there and no luck. Any idea what I'm missing? Thank you

The error I got,

  1. ValueError: Input 0 is incompatible with layer conv1d_1: expected ndim=3, found ndim=2

Code is included below

tb = [keras.callbacks.TensorBoard(log_dir='./dllogs')]

main_input = Input(shape=(294, ), dtype='float32')

x_train = np.array(x_train)
y_train = np.array(y_train)

x_dev = np.array(x_dev)
y_dev = np.array(y_dev)
x = Conv1D(128, 5, activation='relu')(main_input)
x = MaxPooling1D(5)(x)
x = Conv1D(128, 5, activation='relu')(x)
x = MaxPooling1D(5)(x)
x = Conv1D(128, 5, activation='relu')(x)
x = MaxPooling1D(35)(x)
x = Flatten()(x)
x = Dropout(0.25)(x)
x = Dense(128, activation='relu')(x)

preds = Dense(pred_dim, activation='softmax')(x)
model = Model(inputs=main_input, outputs=preds)
model.compile(optimizer='adam', loss='kullback_leibler_divergence', metrics=['accuracy'])
print(model.summary())
history_NN = model.fit(x_train, y_train, batch_size=BATCHSIZE, epochs=EPOCHS, callbacks=tb, validation_data=(x_dev, y_dev))

Answer

  1. The dimensions was an issue that was causing trouble, this was address by the comment by Today.
    1. Next was the pooling dimensions, this is the dimensions reduce after pooling so after the whole thing, it didn't have 35 slides for the final pooling, when I changed it. Fixed.

Resulting code, 

tb = [keras.callbacks.TensorBoard(log_dir='./dllogs')]
main_input = Input(shape=(294, 1))#changed

x_train = np.array(x_train)
y_train = np.array(y_train)

x_dev = np.array(x_dev)
y_dev = np.array(y_dev)
x_train = np.expand_dims(x_train, axis=-1) #changed
x_dev = np.expand_dims(x_dev, axis=-1) #changed
x = Conv1D(128, 5, activation='relu')(main_input)
x = MaxPooling1D(5)(x)
x = Conv1D(128, 5, activation='relu')(x)
x = MaxPooling1D(5)(x)
x = Conv1D(128, 5, activation='relu')(x)
x = MaxPooling1D(11)(x)
x = Flatten()(x)
x = Dropout(0.25)(x)
x = Dense(128, activation='relu')(x)

preds = Dense(pred_dim, activation='softmax')(x)
model = Model(inputs=main_input, outputs=preds)
model.compile(optimizer='adam', loss='kullback_leibler_divergence', metrics=['accuracy'])
print(model.summary())
history_NN = model.fit(x_train, y_train, batch_size=BATCHSIZE, epochs=EPOCHS, callbacks=tb, validation_data=(x_dev, y_dev))

Upvotes: 1

Views: 673

Answers (1)

today
today

Reputation: 33410

Conv1D layer expects sequence inputs of shape (sequence_length, num_features). It seems you have sequences of length 294 with one feature; therefore, each input sample needs to have a shape of (294,1) (and not (294,)). To fix it, you can use np.expand_dims to add a third dimension of size 1 to your input data:

x_train = np.expand_dims(x_train, axis=-1)
x_dev = np.expand_dims(x_dev, axis=-1)

main_input = Input(shape=(294, 1)) # fix the input shape here as well

Upvotes: 1

Related Questions