Github user nltran commented on the pull request:
https://github.com/apache/flink/pull/967#issuecomment-172786594
An ecosystem of 3rd party libraries for Flink makes sense to me. That would
be a good place for experimental features (Ã la Google Labs back in its time)
that are somehow
Github user StephanEwen commented on the pull request:
https://github.com/apache/flink/pull/967#issuecomment-171970520
I think the SSP code look in good shape.
Concerning the parameter server code, I think we should not add this this
to Flink's repository, but it should be hos
Github user hsaputra commented on the pull request:
https://github.com/apache/flink/pull/967#issuecomment-169143768
@nltran Looks like the current state of PR has conflict. Would you mind
rebasing with latest?
Thx!
---
If your project is set up for it, you can reply to this email
Github user nltran commented on the pull request:
https://github.com/apache/flink/pull/967#issuecomment-168965380
Hello guys,
Do you have any plans or updates on this one?
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub a
Github user sachingoel0101 commented on the pull request:
https://github.com/apache/flink/pull/967#issuecomment-138416633
@nltran , I went through you code and if I understand correctly, the
slack-related synchronization is done by the Runtime itself using events. I
like that idea mor
Github user nltran commented on a diff in the pull request:
https://github.com/apache/flink/pull/967#discussion_r38295439
--- Diff:
flink-runtime/src/test/java/org/apache/flink/runtime/iterative/event/EventWithAggregatorsTest.java
---
@@ -123,6 +169,31 @@ private IterationEventWit
Github user tillrohrmann commented on a diff in the pull request:
https://github.com/apache/flink/pull/967#discussion_r37883572
--- Diff:
flink-runtime/src/test/java/org/apache/flink/runtime/iterative/event/EventWithAggregatorsTest.java
---
@@ -123,6 +169,31 @@ private IterationEv
Github user nltran commented on the pull request:
https://github.com/apache/flink/pull/967#issuecomment-132935694
@StephanEwen
Both interfaces for the parameter and the parameter server are quite
similar. The things I see are the following:
- The parameter server code is tight
Github user StephanEwen commented on the pull request:
https://github.com/apache/flink/pull/967#issuecomment-131403208
@nltran How would your use of the parameter server work together with what
is proposed in #1003 ?
---
If your project is set up for it, you can reply to this email a
Github user fhueske commented on the pull request:
https://github.com/apache/flink/pull/967#issuecomment-131030134
Hi @nltran,
thanks for updating the PR!
I don't have time for a thorough review now, but I'll answer your last
question.
ASF homepage states: *["The ASF
Github user nltran commented on the pull request:
https://github.com/apache/flink/pull/967#issuecomment-130734997
The code now passes all the checks :smiley: I have addressed the previous
comments you made previously. @StephanEwen , @fhueske could you give it a
another review? Namely
Github user rmetzger commented on a diff in the pull request:
https://github.com/apache/flink/pull/967#discussion_r36280632
--- Diff:
flink-optimizer/src/main/java/org/apache/flink/optimizer/plantranslate/JobGraphGenerator.java
---
@@ -920,15 +928,33 @@ private JobVertex
createBu
Github user nltran commented on a diff in the pull request:
https://github.com/apache/flink/pull/967#discussion_r36280226
--- Diff:
flink-optimizer/src/main/java/org/apache/flink/optimizer/plantranslate/JobGraphGenerator.java
---
@@ -920,15 +928,33 @@ private JobVertex
createBulk
Github user rmetzger commented on a diff in the pull request:
https://github.com/apache/flink/pull/967#discussion_r36280063
--- Diff:
flink-optimizer/src/main/java/org/apache/flink/optimizer/plantranslate/JobGraphGenerator.java
---
@@ -920,15 +928,33 @@ private JobVertex
createBu
Github user rmetzger commented on a diff in the pull request:
https://github.com/apache/flink/pull/967#discussion_r36280020
--- Diff:
flink-core/src/main/java/org/apache/flink/ps/impl/ParameterServerIgniteImpl.java
---
@@ -0,0 +1,105 @@
+/*
+ * Copyright 2015 EURA NOVA.
Github user rmetzger commented on a diff in the pull request:
https://github.com/apache/flink/pull/967#discussion_r36279991
--- Diff:
flink-core/src/main/java/org/apache/flink/api/common/operators/base/BulkIterationStrategy.java
---
@@ -0,0 +1,30 @@
+/*
+ * Copyright 2015
Github user StephanEwen commented on the pull request:
https://github.com/apache/flink/pull/967#issuecomment-127258102
Ah, my bad. Concerning the initialization, I overlooked the null-check. It
is good the way it is.
---
If your project is set up for it, you can reply to this email a
Github user nltran commented on the pull request:
https://github.com/apache/flink/pull/967#issuecomment-127257276
Thank you for your feedback!
I think this is a fine way to go. I will then provide the parameter server
related code in a different module and include an example o
Github user StephanEwen commented on the pull request:
https://github.com/apache/flink/pull/967#issuecomment-127236312
We are thinking about more general tooling for parameter servers (with
implementations different from Ignite), because the request for that has come
up. Whether we wa
Github user nltran commented on the pull request:
https://github.com/apache/flink/pull/967#issuecomment-127220430
>The code includes Apache Ignite as a parameter server. Is it possible to
make this independent of Ignite? I know that SSP is usually used in parameter
server ML settings,
Github user StephanEwen commented on the pull request:
https://github.com/apache/flink/pull/967#issuecomment-127050940
Wow, impressive contribution! I would very much like to add SSP to Flink.
This is quite a lot, so it may take a bit to review and merge.
Here are some
GitHub user nltran opened a pull request:
https://github.com/apache/flink/pull/967
Stale Synchronous Parallel Iterations
Here is a pull request containing our development on [Stale Synchronous
Parallel
(SSP)](http://reports-archive.adm.cs.cmu.edu/anon/ml2013/CMU-ML-13-103.pdf)
ite
22 matches
Mail list logo
