Add an example for static pipeline with in-memory data and show how to get class probabilities #1953

wschin · 2018-12-22T00:13:41Z

Fixes #1947 and fixes #1881.

TomFinley · 2018-12-23T16:17:57Z

This may be worthwhile as an example, but it would be better if we could create a static pipeline out of a type directly, since the entire purpose of the static pipeline idiom is type safety and compile time checks, which should be achievable, and yet which this does not do. So perhaps a note in the sample that it is not intended to be a permanent solution, and is intended not a serious suggestion forever but more as a stopgap solution until the static pipeline evolves further?

wschin · 2019-01-02T17:14:15Z

@TomFinley, I added your comment to an inline comment. Do you think it's ok now?

sfilipi · 2019-01-02T22:37:40Z

docs/samples/Microsoft.ML.Samples/Static/LightGBMMulticlassWithInMemoryData.cs

+        /// <summary>
+        /// Helper function used to generate <see cref="NativeExample"/>s.
+        /// </summary>
+        private static List<NativeExample> GenerateRandomExamples(int count)


private static List GenerateRandomExamples(int count) [](start = 5, length = 71)

I'd move this in the SampleUtils project, and reference it from the test, and from here.

If we move this to SamplesUtils, then user will not able to see a full example in doc.

In reply to: 244875572 [](ancestors = 244875572)

i believe the main purpose of those examples in the docs are to get copy/pasted in a VS project, add the necessary references and study it there, otherwise int he web page many things don't make sense.

There is also guideline to keep those examples small, and too the point, otherwise their length is an immediate turnoff for average user.

In reply to: 245743767 [](ancestors = 245743767,244875572)

Ok. Will move NativeExample and its generator to SamplesUtils.

In reply to: 245796925 [](ancestors = 245796925,245743767,244875572)

sfilipi · 2019-01-02T22:38:06Z

docs/samples/Microsoft.ML.Samples/Static/LightGBMMulticlassWithInMemoryData.cs

+        {
+            // Create a general context for ML.NET operations. It can be used for exception tracking and logging,
+            // as a catalog of available operations and as the source of randomness.
+            var mlContext = new MLContext(seed: 1, conc: 1);


seed: 1, conc: 1 [](start = 42, length = 16)

seed: 1, conc: 1 [](start = 42, length = 16)

let's omit those in samples, and go with defaults as much as we can. #Closed

Sure.

In reply to: 244875689 [](ancestors = 244875689)

sfilipi · 2019-01-02T22:38:39Z

docs/samples/Microsoft.ML.Samples/Static/LightGBMMulticlassWithInMemoryData.cs

+            var mlContext = new MLContext(seed: 1, conc: 1);
+
+            // Context for calling static classifiers. It contains constructors of classifiers and evaluation utilities.
+            var ctx = new MulticlassClassificationContext(mlContext);


remove this, and use mlContext.MulticlassClassificationContext further down #Closed

sfilipi · 2019-01-02T22:40:19Z

docs/samples/Microsoft.ML.Samples/Static/LightGBMMulticlassWithInMemoryData.cs

+
+            // Show prediction result for the 3rd example.
+            var nativePrediction = nativePredictions[2];
+            Console.WriteLine("Our predicted label to this example is {0} with probability {1}",


Console.WriteLine(" [](start = 11, length = 20)

have a comment for the WriteLines showing here directly what the output would be. #Closed

No problem. Will do the same for Line 159 below.

In reply to: 244876086 [](ancestors = 244876086)

sfilipi · 2019-01-02T22:41:00Z

docs/samples/Microsoft.ML.Samples/Static/LightGBMMulticlassWithInMemoryData.cs

+            return examples;
+        }
+
+        public void MultiClassLightGbmStaticPipelineWithInMemoryData()


MultiClassLightGbmStaticPipelineWithInMemoryData [](start = 20, length = 48)

i think most of the emphasis in 1.0 is on the dynamic API. Would this example add value as a dynamic sample? #WontFix

Maybe in another PR if really needed.

In reply to: 244876215 [](ancestors = 244876215)

sfilipi · 2019-01-02T22:41:41Z

docs/samples/Microsoft.ML.Samples/Static/LightGBMMulticlassWithInMemoryData.cs

+
+namespace Microsoft.ML.Samples.Static
+{
+    class LightGBMMulticlassWithInMemoryData


LightGBMMulticlassWithInMemoryData [](start = 10, length = 34)

this needs to be referenced from some other file, otherwise it won't display in the documentation.

Maybe from the LightGBMStatics catalog, if we have one already? #Closed

I will do it following pattern for MF. Please take a look at LightGbmStaticExtensions.cs in the next iteration.

In reply to: 244876344 [](ancestors = 244876344)

…o get class probabilities

…tion

TomFinley · 2019-01-07T18:13:53Z

using Microsoft.ML.Data;

Header? #ByDesign

Refers to: docs/samples/Microsoft.ML.Samples/Static/LightGBMMulticlassWithInMemoryData.cs:1 in 5aa58c3. [](commit_id = 5aa58c3, deletion_comment = False)

TomFinley

Thanks @wschin , looks generally good. Only thing is the header.

TomFinley · 2019-01-07T18:23:59Z

Yeah, that'll do.

In reply to: 450924472 [](ancestors = 450924472)

wschin · 2019-01-07T23:07:44Z

using Microsoft.ML.Data;

Example files usually do contain a header because they could be encapsulated in another class' doc string.

In reply to: 452028645 [](ancestors = 452028645)

Refers to: docs/samples/Microsoft.ML.Samples/Static/LightGBMMulticlassWithInMemoryData.cs:1 in 5aa58c3. [](commit_id = 5aa58c3, deletion_comment = False)

sfilipi

wschin requested review from TomFinley, Ivanidzo4ka and sfilipi December 22, 2018 00:13

wschin self-assigned this Dec 22, 2018

wschin mentioned this pull request Dec 22, 2018

Multi classification - Probability #1881

Closed

wschin force-pushed the an-example branch from 1ef3e76 to 31d2269 Compare January 2, 2019 17:12

sfilipi reviewed Jan 2, 2019

View reviewed changes

wschin added 4 commits January 7, 2019 09:18

Add an example for static pipeline with in-memory data and show how t…

683bd3b

…o get class probabilities

Really extract labels from learned pipeline

c92c431

Extend a comment to mention that extracting labels is a temporal solu…

56c53f2

…tion

Bypass 32-bit LightGbm test

6fce105

wschin force-pushed the an-example branch from 8380579 to 6fce105 Compare January 7, 2019 17:19

Address comments

5aa58c3

TomFinley approved these changes Jan 7, 2019

View reviewed changes

Move example data structure to SamplesUtils

2f5ed5f

sfilipi approved these changes Jan 8, 2019

View reviewed changes

wschin merged commit 17bdb98 into dotnet:master Jan 8, 2019

wschin deleted the an-example branch January 8, 2019 19:35

ghost locked as resolved and limited conversation to collaborators Mar 25, 2022

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add an example for static pipeline with in-memory data and show how to get class probabilities #1953

Add an example for static pipeline with in-memory data and show how to get class probabilities #1953

wschin commented Dec 22, 2018 •

edited

Loading

TomFinley commented Dec 23, 2018 •

edited

Loading

wschin commented Jan 2, 2019

sfilipi Jan 2, 2019

wschin Jan 7, 2019

sfilipi Jan 7, 2019

wschin Jan 7, 2019

sfilipi Jan 2, 2019 •

edited

Loading

wschin Jan 7, 2019

sfilipi Jan 2, 2019 •

edited

Loading

sfilipi Jan 2, 2019 •

edited

Loading

wschin Jan 7, 2019

sfilipi Jan 2, 2019 •

edited by wschin

Loading

wschin Jan 7, 2019

sfilipi Jan 2, 2019 •

edited

Loading

wschin Jan 7, 2019 •

edited

Loading

TomFinley commented Jan 7, 2019 •

edited by wschin

Loading

TomFinley left a comment

TomFinley commented Jan 7, 2019

wschin commented Jan 7, 2019

sfilipi left a comment

Add an example for static pipeline with in-memory data and show how to get class probabilities #1953

Add an example for static pipeline with in-memory data and show how to get class probabilities #1953

Conversation

wschin commented Dec 22, 2018 • edited Loading

TomFinley commented Dec 23, 2018 • edited Loading

wschin commented Jan 2, 2019

sfilipi Jan 2, 2019

Choose a reason for hiding this comment

wschin Jan 7, 2019

Choose a reason for hiding this comment

sfilipi Jan 7, 2019

Choose a reason for hiding this comment

wschin Jan 7, 2019

Choose a reason for hiding this comment

sfilipi Jan 2, 2019 • edited Loading

Choose a reason for hiding this comment

wschin Jan 7, 2019

Choose a reason for hiding this comment

sfilipi Jan 2, 2019 • edited Loading

Choose a reason for hiding this comment

sfilipi Jan 2, 2019 • edited Loading

Choose a reason for hiding this comment

wschin Jan 7, 2019

Choose a reason for hiding this comment

sfilipi Jan 2, 2019 • edited by wschin Loading

Choose a reason for hiding this comment

wschin Jan 7, 2019

Choose a reason for hiding this comment

sfilipi Jan 2, 2019 • edited Loading

Choose a reason for hiding this comment

wschin Jan 7, 2019 • edited Loading

Choose a reason for hiding this comment

TomFinley commented Jan 7, 2019 • edited by wschin Loading

TomFinley left a comment

Choose a reason for hiding this comment

TomFinley commented Jan 7, 2019

wschin commented Jan 7, 2019

sfilipi left a comment

Choose a reason for hiding this comment

wschin commented Dec 22, 2018 •

edited

Loading

TomFinley commented Dec 23, 2018 •

edited

Loading

sfilipi Jan 2, 2019 •

edited

Loading

sfilipi Jan 2, 2019 •

edited

Loading

sfilipi Jan 2, 2019 •

edited

Loading

sfilipi Jan 2, 2019 •

edited by wschin

Loading

sfilipi Jan 2, 2019 •

edited

Loading

wschin Jan 7, 2019 •

edited

Loading

TomFinley commented Jan 7, 2019 •

edited by wschin

Loading