[ML] Add loss_function to regression by dimitris-athanasiou · Pull Request #56118 · elastic/elasticsearch

dimitris-athanasiou · 2020-05-04T14:31:47Z

Adds parameters loss_function and loss_function_parameter
to regression.

Adds parameters `loss_function` and `loss_function_parameter` to regression.

elasticmachine · 2020-05-04T14:31:49Z

Pinging @elastic/ml-core (:ml)

benwtrent

I have an overall design question.

To my knowledge, only the huber function will actually use the lossFunctionParameter. It almost seems to me that this could be a named object.

If we don't want the complexity a named object requires, what do you think of adding validations? We shouldn't allow the lossFunctionParameter unless it is being used.

dimitris-athanasiou · 2020-05-04T18:55:04Z

msle also uses the parameter. I also had the same concern and I discussed it with @valeriy42. Valeriy is confident we won't be introducing new loss functions in which case it probably makes sense to keep this simple and simply ignore the parameter for mse.

valeriy42 · 2020-05-04T19:49:53Z

@benwtrent you are right, in a canonical form MSLE does not require a parameter, however, @tveasey had a useful insight that log(a+1) in the MSLE formulation actually defines the "area of interest" for the loss function. We are going to change it to log(a+t) to be able to "focus" the loss function where we need it.

lcawl · 2020-05-04T21:17:54Z

docs/reference/ml/df-analytics/apis/put-dfanalytics.asciidoc


+`loss_function`::::
+(Optional, string)
+The loss function used during regression. Available options are `mse` (Mean Squared Error),


I don't think this requires capitalization

Suggested change

The loss function used during regression. Available options are `mse` (Mean Squared Error),

The loss function used during regression. Available options are `mse` (mean squared error),

lcawl · 2020-05-04T21:20:21Z

docs/reference/ml/df-analytics/apis/put-dfanalytics.asciidoc

+`loss_function`::::
+(Optional, string)
+The loss function used during regression. Available options are `mse` (Mean Squared Error),
+`msle` (Mean Squared Logarithmic Error),  `huber` (Pseudo-Huber loss). Defaults to `mse`.


Ditto re mean squared logarithmic, whereas huber seems to be a person's name and can be capitalized

Suggested change

`msle` (Mean Squared Logarithmic Error), `huber` (Pseudo-Huber loss). Defaults to `mse`.

`msle` (mean squared logarithmic error), `huber` (Pseudo-Huber loss). Defaults to `mse`.

lcawl · 2020-05-04T21:21:11Z

docs/reference/ml/df-analytics/apis/put-dfanalytics.asciidoc

+
+`loss_function_parameter`::::
+(Optional, double)
+A strictly positive number that is used as a parameter to the `loss_function`.


A strictly positive number ...

Why is "strictly" required here?

Fair point! I'll remove it.

dimitris-athanasiou · 2020-05-05T08:43:48Z

@elasticmachine update branch

Until elastic#56118 is backported to 7.x.

Until #56118 is backported to 7.x.

Adds parameters `loss_function` and `loss_function_parameter` to regression. Backport of elastic#56118

Adds parameters `loss_function` and `loss_function_parameter` to regression. Backport of #56118

... of elastic#56118

... of #56118

[ML] Add loss_function to regression

a700ff7

Adds parameters `loss_function` and `loss_function_parameter` to regression.

dimitris-athanasiou added >enhancement :ml Machine learning v8.0.0 v7.8.0 labels May 4, 2020

Fix tests

765c905

benwtrent self-requested a review May 4, 2020 17:55

benwtrent reviewed May 4, 2020

View reviewed changes

benwtrent approved these changes May 4, 2020

View reviewed changes

lcawl reviewed May 4, 2020

View reviewed changes

Address some comments about docs

6acb927

Merge branch 'master' into add-loss-function-to-regression

9842a12

dimitris-athanasiou merged commit 6bf3834 into elastic:master May 5, 2020

dimitris-athanasiou deleted the add-loss-function-to-regression branch May 5, 2020 09:36

dimitris-athanasiou added a commit to dimitris-athanasiou/elasticsearch that referenced this pull request May 5, 2020

[ML] Mute data frame analytics BWC tests

60221f8

Until elastic#56118 is backported to 7.x.

dimitris-athanasiou mentioned this pull request May 5, 2020

[ML] Mute data frame analytics BWC tests #56183

Merged

dimitris-athanasiou added a commit that referenced this pull request May 5, 2020

[ML] Mute data frame analytics BWC tests (#56183)

ed0c665

Until #56118 is backported to 7.x.

dimitris-athanasiou added a commit to dimitris-athanasiou/elasticsearch that referenced this pull request May 5, 2020

[7.x][ML] Add loss_function to regression (elastic#56118)

c043ad6

Adds parameters `loss_function` and `loss_function_parameter` to regression. Backport of elastic#56118

dimitris-athanasiou mentioned this pull request May 5, 2020

[7.x][ML] Add loss_function to regression (#56118) #56187

Merged

dimitris-athanasiou added a commit that referenced this pull request May 5, 2020

[7.x][ML] Add loss_function to regression (#56118) (#56187)

75dadb7

Adds parameters `loss_function` and `loss_function_parameter` to regression. Backport of #56118

dimitris-athanasiou mentioned this pull request May 5, 2020

[ML] Adjust BWC versions and unmute tests after backport #56193

Merged

dimitris-athanasiou added a commit to dimitris-athanasiou/elasticsearch that referenced this pull request May 5, 2020

[ML] Adjust BWC versions and unmute tests after backport

5618c32

... of elastic#56118

dimitris-athanasiou added a commit that referenced this pull request May 5, 2020

[ML] Adjust BWC versions and unmute tests after backport (#56193)

d2a7a1e

... of #56118

walterra mentioned this pull request May 6, 2020

[ML] Data Frame Analytics: Skip job cloning functional tests. elastic/kibana#65465

Merged

peteharverson mentioned this pull request May 6, 2020

[ML] Add check for loss_function when cloning a regression job elastic/kibana#65481

Merged

1 task

jakelandis added v8.0.0-alpha1 and removed v8.0.0 labels Jul 26, 2021

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[ML] Add loss_function to regression#56118

[ML] Add loss_function to regression#56118
dimitris-athanasiou merged 4 commits intoelastic:masterfrom
dimitris-athanasiou:add-loss-function-to-regression

dimitris-athanasiou commented May 4, 2020

Uh oh!

elasticmachine commented May 4, 2020

Uh oh!

benwtrent left a comment

Uh oh!

dimitris-athanasiou commented May 4, 2020

Uh oh!

valeriy42 commented May 4, 2020

Uh oh!

lcawl May 4, 2020

Uh oh!

lcawl May 4, 2020

Uh oh!

lcawl May 4, 2020

Uh oh!

dimitris-athanasiou May 5, 2020

Uh oh!

dimitris-athanasiou commented May 5, 2020

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

	The loss function used during regression. Available options are `mse` (Mean Squared Error),
	The loss function used during regression. Available options are `mse` (mean squared error),

	`msle` (Mean Squared Logarithmic Error), `huber` (Pseudo-Huber loss). Defaults to `mse`.
	`msle` (mean squared logarithmic error), `huber` (Pseudo-Huber loss). Defaults to `mse`.

Conversation

dimitris-athanasiou commented May 4, 2020

Uh oh!

elasticmachine commented May 4, 2020

Uh oh!

benwtrent left a comment

Choose a reason for hiding this comment

Uh oh!

dimitris-athanasiou commented May 4, 2020

Uh oh!

valeriy42 commented May 4, 2020

Uh oh!

lcawl May 4, 2020

Choose a reason for hiding this comment

Uh oh!

lcawl May 4, 2020

Choose a reason for hiding this comment

Uh oh!

lcawl May 4, 2020

Choose a reason for hiding this comment

Uh oh!

dimitris-athanasiou May 5, 2020

Choose a reason for hiding this comment

Uh oh!

dimitris-athanasiou commented May 5, 2020

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants