Skip to content

Enhancements to online linear trainers to make them stateless. #1455

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 2 commits into from
Oct 31, 2018
Merged

Enhancements to online linear trainers to make them stateless. #1455

merged 2 commits into from
Oct 31, 2018

Conversation

@TomFinley
Copy link
Contributor

@TomFinley TomFinley commented Oct 30, 2018

  • Factor stateful logic into a separate internal object.
  • Remove direct usage of Console.Writeline
  • Opportunistic fixes of minor issues.

Fixes #1358 .

@TomFinley TomFinley self-assigned this Oct 30, 2018
Zruty0
Zruty0 approved these changes Oct 30, 2018
View reviewed changes
Copy link
Contributor

@Zruty0 Zruty0 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

:shipit:

@sfilipi
Copy link
Member

sfilipi commented Oct 30, 2018 •
edited
Loading 

    public void OnlineLinearWorkout()

might be worth to add a test involving CV. I run into this using CV:

#1358

#Resolved

Refers to: test/Microsoft.ML.Tests/TrainerEstimators/OnlineLinearTests.cs:16 in e811fb1. [](commit_id = e811fb1, deletion_comment = False)

@sfilipi sfilipi mentioned this pull request Oct 30, 2018
Closed
@TomFinley
Copy link
Contributor Author

TomFinley commented Oct 30, 2018 •
edited
Loading 

    public void OnlineLinearWorkout()

Hi @sfilipi, while it's true that's how the issue was discovered, I kind of feel like the issue of whether IEstimator can be callable multiple times and produce sensible results ought to be handled in a more systematic way than just remembering "also write a CV test" for every ITrainerEstimator implementation.

I think once we actually do switch the CV tests do use the CV API vs. the command line this will just happen naturally.

In reply to: 434448974 [](ancestors = 434448974)

Refers to: test/Microsoft.ML.Tests/TrainerEstimators/OnlineLinearTests.cs:16 in e811fb1. [](commit_id = e811fb1, deletion_comment = False)

@TomFinley TomFinley force-pushed the tfinley/LinearState branch from e811fb1 to 7d713a3 Compare October 30, 2018 21:05
sfilipi
sfilipi approved these changes Oct 30, 2018
View reviewed changes
Copy link
Member

@sfilipi sfilipi left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

:shipit:

sfilipi
sfilipi reviewed Oct 30, 2018
View reviewed changes
src/Microsoft.ML.StandardLearners/Standard/Online/OnlineLearnerCatalog.cs
var env = CatalogUtils.GetEnvironment(ctx);
var loss = new TrivialClassificationLossFactory(lossFunction ?? new LogLoss());
return new AveragedPerceptronTrainer(env, label, features, weights, loss, learningRate, decreaseLearningRate, l2RegularizerWeight, numIterations, advancedSettings);
return new AveragedPerceptronTrainer(env, label, features, weights, lossFunction ?? new LogLoss(), learningRate, decreaseLearningRate, l2RegularizerWeight, numIterations, advancedSettings);
Copy link
Member

@sfilipi sfilipi Oct 30, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LogLoss() [](start = 100, length = 9)

it's LogLoss here, but the default for AP is HingeLoss.

@TomFinley
Enhancements to online linear trainers to make them stateless.
fb09a0b 
* Factor stateful logic into a separate internal object.
* Remove direct usage of Console.Writeline
* Opportunistic fixes of minor issues.
@TomFinley
Nuke failing Mac test
9c173c8
@TomFinley TomFinley force-pushed the tfinley/LinearState branch from 5ad6f06 to 9c173c8 Compare October 30, 2018 23:49
@TomFinley TomFinley merged commit 696d511 into dotnet:master Oct 31, 2018
@TomFinley TomFinley deleted the tfinley/LinearState branch November 3, 2018 09:10
@antoniovs1029 antoniovs1029 mentioned this pull request Jan 31, 2020
Closed
@ghost ghost locked as resolved and limited conversation to collaborators Mar 27, 2022
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.

Reviewers

@Ivanidzo4ka Ivanidzo4ka Awaiting requested review from Ivanidzo4ka

2 more reviewers

@sfilipi sfilipi sfilipi approved these changes

@Zruty0 Zruty0 Zruty0 approved these changes

Reviewers whose approvals may not affect merge requirements

Assignees

@TomFinley TomFinley

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

"Can't re-use trainer!" message on CV with AveragePerceptron.

3 participants

@TomFinley @sfilipi @Zruty0