Loading TOC...

cntk:learning-rate-schedule-from-sequence

cntk:learning-rate-schedule-from-sequence(
   $learning-rates as xs:double*,
   [$epoch-size as xs:unsignedLong],
   [$minibatch-size as xs:unsignedLong]
) as cntk:schedule

Summary

Returns a learning rate schedule, based on a sequence of learning rates. For epoch $i, the learning rate will be set to $learning-rates[$i]. If the number of epochs is larger the the length of $learning-rates, the last learning rate will be used for the rest of epochs.

Parameters
$learning-rates A sequence of learning rates.
$epoch-size The epoch size. The learning rate changes every $epoch-size samples. Defaults to a special sentinel value to represent the whole data sweep.
$minibatch-size The minibatch size. Defaults to a sentinel value which means ignored. This is a "best-effort" parameter: there is no guarantee that the size of every minibatch is exactly $minibatch-size.

Example

  cntk:learning-rate-schedule-from-sequence((0.1,0.05,0.01), 100)
  => a cntk:schedule, where the first 100 samples are using 0.1 as
learning rate, the second 100 samples using 0.05, and all subsequence samples
use 0.01.

Stack Overflow iconStack Overflow: Get the most useful answers to questions from the MarkLogic community, or ask your own question.