Github user avulanov commented on the pull request:
https://github.com/apache/spark/pull/1290#issuecomment-99289880
@hhbyyh The new version allocates memory needed for storing model
parameters and intermediate results only once and then reuses it. The actual
speed should be up to several times faster, depending on the batch size and
number of iterations. The time of one iteration is comparable to Caffe with the
same settings, because most of the time is spent in BLAS dgemm (matrix-matrix
multiplication). Also, new version has much less probability of hitting the
garbage collector. Probably I will plot some comparisons later.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
