CODEWITHSUNDEEP
pythontensorflowkerasactivation-function
Pavithran Ravichandiran
Pavithran Ravichandiran

Reputation: 1891

Advanced Custom activation function in keras + tensorflow

def newactivation(x):
    if x>0:
        return K.relu(x, alpha=0, max_value=None)
    else :
        return x * K.sigmoid(0.7* x)

get_custom_objects().update({'newactivation': Activation(newactivation)})

I am trying to use this activation function for my model in keras, but I am having hard time by finding what to replace 

if x>0:

ERROR i got:

File "/usr/local/lib/python3.4/dist-packages/tensorflow/python/framework/ops.py", line 614, in bool raise TypeError("Using a tf.Tensor as a Python bool is not allowed. "

TypeError: Using a tf.Tensor as a Python bool is not allowed. Use if >t is not None: instead of if t: to test if a tensor is defined, and >use TensorFlow ops such as tf.cond to execute subgraphs conditioned on >the value of a tensor.

Can someone make it clear for me?

Upvotes: 2

Views: 2481

Answers (4)

Robin
Robin

Reputation: 3

inspired by the previous answer from ed Mar 21 '18 at 17:28 tomhosking. This worked for me. tf.cond

def custom_activation(x):
    return tf.cond(tf.greater(x, 0), lambda: ..., lambda: ....)

Upvotes: 0

user239457
user239457

Reputation: 1876

You can evaluate the tensor and then check for the condition

from keras.backend.tensorflow_backend import get_session


sess=get_session()
if sess.run(x)>0:
    return t1
else:
    return t2

get_session is not available for TensorFlow 2.0. Solution for that you can find here

Upvotes: 0

tomhosking
tomhosking

Reputation: 61

Try something like:

def newactivation(x):
    return tf.cond(x>0, x, x * tf.sigmoid(0.7* x))

x isn't a python variable, it's a Tensor that will hold a value when the model is run. The value of x is only known when that op is evaluated, so the condition needs to be evaluated by TensorFlow (or Keras).

Upvotes: 3

Jakub Bartczuk
Jakub Bartczuk

Reputation: 2378

if x > 0 doesn't make sense because x > 0 is a tensor, and not a boolean value.

To do a conditional statement in Keras use keras.backend.switch.

For example your 

if x > 0:
   return t1
else:
   return t2

Would become 

keras.backend.switch(x > 0, t1, t2)

Upvotes: 6

Related Questions