Keras ValueError: Unknown activation function:<lambda>

Question

I'm using a lambda function to create a custom activation function but when I try to upload a checkpoint I get the error:

ValueError: Unknown activation function:<lambda>

The function is:

lrelu = lambda x: tf.keras.activations.relu(x, alpha=0.2)

and used like this:

Conv2D(filters=96, kernel_size=(3,3),strides=(2,2),activation=lrelu)

I tried to add the custom_objects with no luck:

model = load_model(filepath, custom_objects = {"lrelu": lrelu})

I know I can replace my function with an extra layer and avoid this problem, but I was wondering if there is a way to make this work.

Thanks.

OverLordGoldDragon · Accepted Answer · 2019-12-07 18:49:08Z

2

You'll need the Lambda layer wrapper - minimal example below.

from keras.layers import Input, Conv2D, Lambda
from keras.models import Model
import tensorflow as tf
import numpy as np

lrelu = Lambda(lambda x: tf.keras.activations.relu(x, alpha=0.2))

ipt   = Input((4,4,3))
out   = Conv2D(3, 1, activation=lrelu)(ipt)
model = Model(ipt, out)
model.compile('adam', 'mse')

x = np.random.randn(32,4,4,3)
model.fit(x, x)

32/32 [==============================] - 6s 201ms/sample - loss: 2.1475

edited Dec 7, 2019 at 18:49

answered Nov 24, 2019 at 18:08

OverLordGoldDragon

20k12 gold badges60 silver badges121 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

joaquin7 Over a year ago

Unfortunately, this is using tensorflow.keras which is not the same as just keras. And because I'm using a custom implementation I cannot switch

OverLordGoldDragon Over a year ago

@joaquin7 It still works - and in the future, best to specify your import sources (e.g. from keras)

leenremm · Accepted Answer · 2019-11-29 15:38:52Z

0

Unfortunately it seems that this is an issue with Keras.

Work-around is to add a separate Activation Layer:

model.add(Dense(units=128))
model.add(Lambda(lambda x: custom_activation(x, salience_array[0])))

Note: To save and load a model, use Lambda() instead of Activation().

Reference: https://github.com/keras-team/keras/issues/8880

answered Nov 29, 2019 at 15:38

leenremm

1,47418 silver badges22 bronze badges

Collectives™ on Stack Overflow

Keras ValueError: Unknown activation function:<lambda>

2 Answers 2

2 Comments

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

2 Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Related