CODEWITHSUNDEEP
pythontensorflowkeras
ikenas
ikenas

Reputation: 431

How to migrate Dense layers from Tensorflow 1 to Tensorflow 2?

How I could migrate this layer to tf2

observations = tf.placeholder(tf.float32,[None, OBSERVATIONS_SIZE])

h = tf.layers.dense(
     observations,
     units=hidden_layer_size,
     activation=tf.nn.relu,
     kernel_initializer=tf.contrib.layers.xavier_initializer()
)

I Found that the placeholder now is 'Input' and I used the Dense layers for tf2

I tried with:

observations = tf.keras.Input(
    shape = [ None, OBSERVATIONS_SIZE ],
    dtype = tf.float32
)

h = tf.keras.layers.Dense(
     observations,
     units=hidden_layer_size,
     activation='relu',
     kernel_initializer = 'glorot_uniform'
)

I get this error if i use it

TypeError: __init__() got multiple values for argument 'units'

How i should use the placeholder/Input in this case?

Upvotes: 1

Views: 243

Answers (1)

Dr. Snoopy
Dr. Snoopy

Reputation: 56377

Keras layers are not used as tf.layers, they are callable instead of passing a tensor as the first parameter, so it should be:

observations = tf.keras.Input(
    shape = [ None, OBSERVATIONS_SIZE ],
    dtype = tf.float32
    )

h = tf.keras.layers.Dense(
     units=hidden_layer_size,
     activation='relu',
     kernel_initializer = 'glorot_uniform'
     )(observations)

Upvotes: 2

Related Questions