cjolivier01 commented on issue #9366: Refactor dropout operator to use
ParallelRandom generator and also react deterministically when seeding
URL: https://github.com/apache/incubator-mxnet/pull/9366#issuecomment-356459718
How do you propose?
cjolivier01 commented on issue #9366: Refactor dropout operator to use
ParallelRandom generator and also react deterministically when seeding
URL: https://github.com/apache/incubator-mxnet/pull/9366#issuecomment-356645384
Ok, I will add some unit tests with edge cases and to make sure sums
cjolivier01 commented on issue #9366: Refactor dropout operator to use
ParallelRandom generator and also react deterministically when seeding
URL: https://github.com/apache/incubator-mxnet/pull/9366#issuecomment-356848301
done ready to merge
cjolivier01 commented on issue #9366: Refactor dropout operator to use
ParallelRandom generator and also react deterministically when seeding
URL: https://github.com/apache/incubator-mxnet/pull/9366#issuecomment-357087320
Do you mean compared to the previous MKL implementation?
On Th
cjolivier01 commented on issue #9366: Refactor dropout operator to use
ParallelRandom generator and also react deterministically when seeding
URL: https://github.com/apache/incubator-mxnet/pull/9366#issuecomment-357107778
MKL seems to be quite a bit faster, tryinbg to see how to reenable it
cjolivier01 commented on issue #9366: Refactor dropout operator to use
ParallelRandom generator and also react deterministically when seeding
URL: https://github.com/apache/incubator-mxnet/pull/9366#issuecomment-357107778
MKL seems to be quite a bit faster, trying to see how to reenable it
cjolivier01 commented on issue #9366: Refactor dropout operator to use
ParallelRandom generator and also react deterministically when seeding
URL: https://github.com/apache/incubator-mxnet/pull/9366#issuecomment-357121396
It is so much faster on MKL that we might want to consider one day us
cjolivier01 commented on issue #9366: Refactor dropout operator to use
ParallelRandom generator and also react deterministically when seeding
URL: https://github.com/apache/incubator-mxnet/pull/9366#issuecomment-357253246
you can?t pass a CPU pointer as an argument to a GPU function because
cjolivier01 commented on issue #9366: Refactor dropout operator to use
ParallelRandom generator and also react deterministically when seeding
URL: https://github.com/apache/incubator-mxnet/pull/9366#issuecomment-357328496
It is, at least for a given machine and a given OMP configuration. I
cjolivier01 commented on issue #9366: Refactor dropout operator to use
ParallelRandom generator and also react deterministically when seeding
URL: https://github.com/apache/incubator-mxnet/pull/9366#issuecomment-357328496
It is, at least for a given machine and a given OMP configuration. I
cjolivier01 commented on issue #9366: Refactor dropout operator to use
ParallelRandom generator and also react deterministically when seeding
URL: https://github.com/apache/incubator-mxnet/pull/9366#issuecomment-357386015
Unless I was reading something wrong before this would blow up for an
cjolivier01 commented on issue #9366: Refactor dropout operator to use
ParallelRandom generator and also react deterministically when seeding
URL: https://github.com/apache/incubator-mxnet/pull/9366#issuecomment-357386015
Unless I was reading something wrong before this would blow up for an
cjolivier01 commented on issue #9366: Refactor dropout operator to use
ParallelRandom generator and also react deterministically when seeding
URL: https://github.com/apache/incubator-mxnet/pull/9366#issuecomment-357386015
Unless I was reading something wrong before this would blow up for an
cjolivier01 commented on issue #9366: Refactor dropout operator to use
ParallelRandom generator and also react deterministically when seeding
URL: https://github.com/apache/incubator-mxnet/pull/9366#issuecomment-357489640
keep in mind that this just generates a bernoulli distribution and no
cjolivier01 commented on issue #9366: Refactor dropout operator to use
ParallelRandom generator and also react deterministically when seeding
URL: https://github.com/apache/incubator-mxnet/pull/9366#issuecomment-35259
actually i misspoke before ? this just generates bernoulli distributi
cjolivier01 commented on issue #9366: Refactor dropout operator to use
ParallelRandom generator and also react deterministically when seeding
URL: https://github.com/apache/incubator-mxnet/pull/9366#issuecomment-357814840
ok rearranged a little. removed the switch.
Added comment
---
cjolivier01 commented on issue #9366: Refactor dropout operator to use
ParallelRandom generator and also react deterministically when seeding
URL: https://github.com/apache/incubator-mxnet/pull/9366#issuecomment-35259
actually i misspoke before ? this just generates bernoulli distributi
cjolivier01 commented on issue #9366: Refactor dropout operator to use
ParallelRandom generator and also react deterministically when seeding
URL: https://github.com/apache/incubator-mxnet/pull/9366#issuecomment-358055283
Ok, if no has an objection, I plan to merge this this afternoon.
---
cjolivier01 commented on issue #9366: Refactor dropout operator to use
ParallelRandom generator and also react deterministically when seeding
URL: https://github.com/apache/incubator-mxnet/pull/9366#issuecomment-358059834
I have added JIRA: https://issues.apache.org/jira/browse/MXNET-4
19 matches
Mail list logo
