[SPARK-9906] [ML] User guide for LogisticRegressionSummary #8197

MechCoder · 2015-08-14T11:21:33Z

User guide for LogisticRegression summaries

MechCoder · 2015-08-14T11:22:37Z

docs/ml-guide.md

@feynmanliang Is there are a shorthand to do this directly in Spark SQL? Once I understand that I can update the Java Example as well.

You might be able to get it done by doing an aggregation (e.g. max) but I haven't tried myself. In any case, I think just keeping everything up to L868 is sufficient since L870-872 aren't really showing how to use the summary feature anyways

I had thought the user can choose the best threshold and re-run LogisticRegression on this best threshold.

In the first run, it will be a random sampling of the dataset and in the second run it will be the entire dataset. Do you still think it will not be useful?

The closest I could come up with is.

val maxFMeasure = fMeasure.select(max(df("F-Measure"))).collect()(0).getFloat(0) val threshold = fMeasure.filter(df("F-Measure") >= maxFMeasure).collect()(0).getFloat(0)

Nice!

I think what you described is useful, but is outside the scope of LogisticRegressionSummary. L869-L872 don't demonstrate any of the functionality these docs are intended to describe, which is why I propose we remove it. What do you think?

OK. I wouldn't complain because it makes my job easier.

SparkQA · 2015-08-14T11:42:53Z

Test build #40866 has finished for PR 8197 at commit 487b361.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

feynmanliang · 2015-08-14T17:00:23Z

docs/ml-guide.md

"toyData" -> "toy data"

The Scala / Java effect :P

SparkQA · 2015-08-15T15:59:58Z

Test build #40966 has finished for PR 8197 at commit 56cb35b.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

feynmanliang · 2015-08-15T23:08:11Z

docs/ml-guide.md

"objectiveHistory and metric" (surround code with backticks so docs apply correct styles to it)

Actually, metrics is not part of the public API

MechCoder · 2015-08-16T04:58:09Z

@feynmanliang ok, I've addressed your comments, anything else?

MechCoder · 2015-08-16T05:04:32Z

retest this please

feynmanliang · 2015-08-16T05:07:27Z

Jenkins test this please

feynmanliang · 2015-08-16T05:07:54Z

Lgtm pending tests

@mengxr

SparkQA · 2015-08-16T11:06:40Z

Test build #40988 has finished for PR 8197 at commit 5244459.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

jkbradley · 2015-08-18T00:57:02Z

I think this should be part of the linear methods section, not part of the main ML guide. Once we have summary support in more model types, then we can mention it in the main ML guide. Could you please merge it with ml-linear-methods.md? Feel free to reorganize and stomp on existing examples as needed since that section is very basic right now. Thanks!

feynmanliang · 2015-08-18T01:01:14Z

docs/ml-guide.md

Since this links the scala api doc, move this under the scala codetab and add another one for the java api doc

feynmanliang · 2015-08-18T01:11:28Z

👍 to @jkbradley 's suggestion; my apologies for not thinking of that and the misleading JIRA task description. I've updated the JIRA to reflect this.

feynmanliang · 2015-08-18T01:13:35Z

@MechCoder I am working on LinearRegressionSummary docs (SPARK-9905) and I think we should communicate about where these docs should go. What do you think about adding a ### Model Summaries section with Linear Regression and Logistic Regression subsections right before the ### Optimization section in ml-linear-methods?

jkbradley · 2015-08-18T01:22:47Z

@feynmanliang That organization sounds good, if the summary examples can make use of previous basic examples (and avoid copying code). Another option would be to intersperse text and code, as in [http://spark.apache.org/docs/latest/mllib-clustering.html#streaming-k-means], but that will make sense only if clear headers can go within the codetabs text.

Documentation cleanup in `ml-linear-methods`

MechCoder · 2015-08-18T18:27:55Z

I have merged your changes. Thanks!

feynmanliang · 2015-08-18T18:29:54Z

Cool LGTM, @jkbradley for final pass

MechCoder · 2015-08-18T18:29:57Z

docs/ml-linear-methods.md

Whoops, yep that's right

SparkQA · 2015-08-18T18:47:20Z

Test build #41151 has finished for PR 8197 at commit 83d229f.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

MechCoder · 2015-08-18T19:19:11Z

@feynmanliang Actually I'm not very convinced that coupling these with existing code is the way to go. We could have a docs/ml-summary.md in the future with subsections about each summary with an example. This would prevent repetition of certain common descriptions across all summaries.

But for now I think this should be okay.

feynmanliang · 2015-08-18T20:01:57Z

Having a ml-summary.md would lead to quite a bit of repeated documentation; we would have to supply model training example code in both the model's description page as well as in ml-summary since the model itself usually cannot be directly instantiated.

What things do you think will be repeated?

MechCoder · 2015-08-18T20:30:09Z

You are right.

But I thought that the example code such as extracting the training loss, and common metrics would be repeated across some classification and regression models.

We could maybe just have one complete example of a classification and regression model with the summaries (the training being one that is not already in the description page) and for the others which can instantiate the summary directly to show how else it can be used apart from the common stuff.

[Just a suggestion and can be thought about later]

SparkQA · 2015-08-18T20:55:33Z

Test build #41162 has finished for PR 8197 at commit 7bf922c.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

MechCoder · 2015-08-18T21:04:31Z

Okay. But he had to told something else in that PR discussion :p. I do
agree that doing model.binarySUmmary is much nearer than
model.asInstanceOf[]…
On Aug 19, 2015 2:27 AM, "UCB AMPLab" [email protected] wrote:

Merged build finished. Test PASSed.

—
Reply to this email directly or view it on GitHub
#8197 (comment).

feynmanliang · 2015-08-25T02:13:48Z

LGTM

CC @jkbradley this is blocking SPARK-9905 so do you mind reviewing when you have a chance? Thanks!

jkbradley · 2015-08-25T17:54:44Z

docs/ml-linear-methods.md

I don't think you need to put lambda in, but if you do, then how about putting it outside of big parentheses or brackets to make the equation easier to read?

jkbradley · 2015-08-25T17:55:26Z

Done with review

jkbradley · 2015-08-27T22:32:51Z

I'll go ahead and merge this and ask @feynmanliang to make any needed updates when he sends a PR for the linear regression summary docs. Thanks!

jkbradley · 2015-08-27T22:32:57Z

merging with master and branch-1.5

User guide for LogisticRegression summaries Author: MechCoder <[email protected]> Author: Manoj Kumar <[email protected]> Author: Feynman Liang <[email protected]> Closes #8197 from MechCoder/log_summary_user_guide. (cherry picked from commit c94ecdf) Signed-off-by: Joseph K. Bradley <[email protected]>

* Adds user guide for `LinearRegressionSummary` * Fixes unresolved issues in #8197 CC jkbradley mengxr Author: Feynman Liang <[email protected]> Closes #8491 from feynmanliang/SPARK-9905. (cherry picked from commit af0e124) Signed-off-by: Xiangrui Meng <[email protected]>

* Adds user guide for `LinearRegressionSummary` * Fixes unresolved issues in #8197 CC jkbradley mengxr Author: Feynman Liang <[email protected]> Closes #8491 from feynmanliang/SPARK-9905.

[SPARK-9906] [ML] User guide for LogisticRegressionSummary

487b361

MechCoder reviewed Aug 14, 2015
View reviewed changes

feynmanliang reviewed Aug 14, 2015
View reviewed changes

minor

56cb35b

MechCoder force-pushed the log_summary_user_guide branch from 7cc3f58 to 56cb35b Compare August 15, 2015 15:42

feynmanliang reviewed Aug 15, 2015
View reviewed changes

remove threshold selection

9831270

MechCoder force-pushed the log_summary_user_guide branch from b79d780 to 5244459 Compare August 16, 2015 10:46

Add example based on threshold selection

1ab3d9c

MechCoder force-pushed the log_summary_user_guide branch from 5244459 to 1ab3d9c Compare August 16, 2015 10:47

feynmanliang reviewed Aug 18, 2015
View reviewed changes

Merge pull request #1 from feynmanliang/log_summary_user_guide

83d229f

Documentation cleanup in `ml-linear-methods`

MechCoder reviewed Aug 18, 2015
View reviewed changes

typo

7bf922c

jkbradley reviewed Aug 25, 2015
View reviewed changes

asfgit closed this in c94ecdf Aug 27, 2015

MechCoder deleted the log_summary_user_guide branch August 27, 2015 23:35

feynmanliang mentioned this pull request Aug 28, 2015

[SPARK-9905][ML][Doc] Adds LinearRegressionSummary user guide #8491

Closed

sethah mentioned this pull request Oct 13, 2016

[SPARK-17139][ML] Add model summary for MultinomialLogisticRegression #15435

Closed

[SPARK-9906] [ML] User guide for LogisticRegressionSummary #8197

[SPARK-9906] [ML] User guide for LogisticRegressionSummary #8197

Uh oh!

Conversation

MechCoder commented Aug 14, 2015

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

SparkQA commented Aug 14, 2015

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

SparkQA commented Aug 15, 2015

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

MechCoder commented Aug 16, 2015

Uh oh!

MechCoder commented Aug 16, 2015

Uh oh!

feynmanliang commented Aug 16, 2015

Uh oh!

feynmanliang commented Aug 16, 2015

Uh oh!

SparkQA commented Aug 16, 2015

Uh oh!

jkbradley commented Aug 18, 2015

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

feynmanliang commented Aug 18, 2015

Uh oh!

feynmanliang commented Aug 18, 2015

Uh oh!

jkbradley commented Aug 18, 2015

Uh oh!

MechCoder commented Aug 18, 2015

Uh oh!

feynmanliang commented Aug 18, 2015

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

SparkQA commented Aug 18, 2015

Uh oh!

MechCoder commented Aug 18, 2015

Uh oh!

feynmanliang commented Aug 18, 2015

Uh oh!

MechCoder commented Aug 18, 2015

Uh oh!

SparkQA commented Aug 18, 2015

Uh oh!

MechCoder commented Aug 18, 2015

Uh oh!

feynmanliang commented Aug 25, 2015

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jkbradley commented Aug 25, 2015

Uh oh!

jkbradley commented Aug 27, 2015

Uh oh!