CODEWITHSUNDEEP
pythontensorflowkerasconv-neural-network
Bruno Eigenmann
Bruno Eigenmann

Reputation: 370

Keras throwing error: ('Keyword argument not understood:', 'init') and ('Keyword argument not understood:', 'dim_ordering')

I've built the following model with Keras from Tensorflow (version = 2.2.4-tf)

model = tf.keras.Sequential()
model.add(Convolution2D(24, 5, 5, padding='same',init='he_normal', input_shape = (target_Width,target_Height, 3),dim_ordering="tf"))
model.add(Activation('relu'))
model.add(GlobalAveragePooling2D())
model.add(Dense(18))

But somehow I'm getting the following error: ('Keyword argument not understood:', 'init') and ('Keyword argument not understood:', 'dim_ordering')

Upvotes: 0

Views: 4650

Answers (2)

desertnaut
desertnaut

Reputation: 60321

None of these arguments exist, according to the documentation.

  • By init, I take it you meant to use kernel_initializer

  • dim_ordering="tf" is used only in stand-alone Keras (and not in tf.keras), as stand-alone Keras can be used with either Tensorflow or Theano backends; these backends use different ordering scheme, hence the need to clarify this. It is not necessary here.

So, here your convolutional layer should be:

model.add(Conv2D(24, 5, 5, padding='same', kernel_initializer='he_normal', input_shape = (target_Width,target_Height, 3)))

If, as the answer answer suggests, you are indeed using keras.layers.convolutional.Convolution2D from the stand-alone Keras package in a tf.keras.Sequential() model, this mixing is highly not recommended, and you should revert to tf.keras.layers.Conv2D instead (and do the same for the rest of the model layers as well).

Upvotes: 2

yudhiesh
yudhiesh

Reputation: 6799

It seems that you are trying to use keras.layers.convolutional.Convolution2D instead of tf.keras.layers.Conv2D. If that is the case then use this instead:

from keras.models import Sequential
model = Sequential()
model.add(keras.layers.convolutional.Convolution2D(24, 5, 5, padding='same',init='he_normal', input_shape = (target_Width,target_Height, 3),dim_ordering="tf"))

Or using Conv2D from tf.keras which does not have the arguments init and dim_ordering:

model = tf.keras.Sequential()
model.add(Conv2D(24, 5, 5, padding='same', kernel_initializer='he_normal', input_shape = (target_Width,target_Height, 3)))

Upvotes: 3

Related Questions