CODEWITHSUNDEEP
tensorflow
Oğuz Şerbetci
Oğuz Şerbetci

Reputation: 2067

Is tf.train.GradientDescentOptimizer a vanilla GradientDescent?

Is tf.train.GradientDescentOptimizer a vanilla gradient descent? i.e. not SGD, so it is equivalent to a gradient update implemented in numpy.

Upvotes: 0

Views: 430

Answers (1)

Shamane Siriwardhana
Shamane Siriwardhana

Reputation: 4201

Yes, it's the vanilla gradient descent. You can't say it is not SGD. Because it depends on your number of examples you consider when training a single epoch.

Which means if you use only a mini-batch (ideally one single instance, but mini-batch is also fine) of data for a single epoch, We call it SGD.

Yes functionally it should be equivalent to numpy implementation.

Upvotes: 1

Related Questions