CODEWITHSUNDEEP
tensorflowmachine-learningkerasdeep-learning
vasu
vasu

Reputation: 183

Is there a way to use global_step in Keras framework?

I am trying to reproduce polynomial decay for learning rate decay in Keras framework which as implemented in Tensorflow framework is given below.

def poly_decay(step, initial_value, decay_period_images_seen):
    """
    Decays a variable using a polynomial law.
    :param step: number of images seen by the network since the beginning of the training.
    :param initial_value: The initial value of the variable to decay..
    :param decay_period_images_seen: the decay period in terms of images seen by the network
    (1 epoch of 10 batches of 6 images each means that 1 epoch = 60 images seen).
    Thus this value must be a multiple of the number of batches
    :return: The decayed variable.
    """

    factor = 1.0 - (tf.cast(step, tf.float32) / float(decay_period_images_seen))
    lrate = initial_value * np.power(factor, 0.9)

    return lrate

Does Keras offer any hidden parameter (which perhaps I don't know about) for global step or is there an equivalent of global step in Keras? Or is there any alternative way to implement polynomial learning rate decay in Keras framework ?

Upvotes: 2

Views: 1439

Answers (1)

ASHu2
ASHu2

Reputation: 2047

Basically, the parameter is itself provided as arguments for optimisers.

Take a look at optimizers.

sgd = optimizers.SGD(lr=0.01, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(loss='mean_squared_error', optimizer=sgd)

So here, you can just pass in the poly_decay() as a parameter.

Usually we use time-based decay instead of polynomial decay:

learning_rate = 0.1
decay_rate = learning_rate / epochs
momentum = 0.8
sgd = SGD(lr=learning_rate, momentum=momentum, decay=decay_rate, nesterov=False)

Check this blog for more reference!!

Upvotes: 1

Related Questions