Activation function as string about probability HOT 3 CLOSED

Looking further at the examples, I think perhaps I should just be using tf.placeholders instead of an InputLayer anyway? But then I still get this error if I do, e.g.,

import tensorflow as tf
import tensorflow_probability as tfp

inputs = tf.placeholder(tf.float32, shape=(None, 10, 10))
layer = tfp.layers.Convolution1DFlipout(32, kernel_size=3, activation='relu')
model = tf.keras.Sequential()
model.add(input_layer)
model.add(layer)
output = model(inputs)

from probability.

@dustinvtran

Hi! I believe activation needs to be a callable. In this case, you should pass in tf.nn.relu.

from probability.

Yes, it requires a callable. Looking at tf.keras.layers.Dense's implementation, strings should definitely be supported.

Contributions are welcome; the change is as easy as tf.keras.layers.Dense's version.

re:input layer. You can still use it.

from probability.