Field-aware factorization machine to estimator #912

sfilipi · 2018-09-14T06:39:25Z

FAFM now extends TrainerEstimatorBase

FAFM doesn't inherit from TrainerEstimatorBase, just implements ITrainerEstimator

Zruty0 · 2018-09-18T03:27:16Z

src/Microsoft.ML.Data/Scorers/PredictionTransformer.cs

@@ -30,23 +32,33 @@ public abstract class PredictionTransformerBase<TModel> : IPredictionTransformer
        protected readonly ISchemaBindableMapper BindableMapper;
        protected readonly ISchema TrainSchema;

-        public string FeatureColumn { get; }
+        public string[] FeatureColumn { get; }


FeatureColumn [](start = 24, length = 13)

oh no... I don' like this change already.

Forcing ALL predictors to expose parallel arrays of feature columns is not a great change #Closed

I don't think there's much overhead from it. Do you think it will cause problems? #Closed

Zruty0 · 2018-09-18T03:29:11Z

test/Microsoft.ML.Tests/TrainerEstimators/FAFMEstimator.cs

+
+            IEstimator<ITransformer> est = new FieldAwareFactorizationMachineTrainer(Env, "Label", new[] { "Feature1", "Feature2", "Feature3", "Feature4" });
+
+            //var result = est.Fit(data);


// [](start = 12, length = 2)

remove commented out code #Resolved

Zruty0 · 2018-09-18T03:30:21Z

test/Microsoft.ML.Tests/TrainerEstimators/OnlineLinearTests.cs

 {
-    public sealed class OnlineLinearTests : TestDataPipeBase
+    public partial class TrainerEstimators


TrainerEstimators [](start = 25, length = 17)

thanks for making this change #ByDesign

Zruty0 · 2018-09-18T03:31:40Z

src/Microsoft.ML.Data/Prediction/IPredictionTransformer.cs

-using Microsoft.ML.Runtime.Internal.Calibration;
-using System;
-using System.Collections.Generic;
-using System.Text;

 namespace Microsoft.ML.Runtime
 {
    public interface IPredictionTransformer<out TModel> : ITransformer


IPredictionTransformer [](start = 21, length = 22)

I don't think we even need an interface for FFM, since for the time being it's the only trainer that accepts multiple feature columns. #Closed

If we want to inquire about all trainers, it is useful to have them extend one interface. #Closed

Zruty0

🕐

…tures. Splitting IPredictionTransformer into two interfaces Creating a transformer wrapping the FAFM predictor.

sfilipi · 2018-09-18T20:58:23Z

test/Microsoft.ML.Tests/TrainerEstimators/FAFMEstimator.cs

+{
+    public partial class TrainerEstimators : TestDataPipeBase
+    {
+        [Fact]


[Fact] [](start = 7, length = 7)

The test is failing to check whether the input is valid for fit. Resolve before check in. #Resolved

Thanks @Zruty0 for fixing the mismatch between the active and inactive columns.

In reply to: 218592628 [](ancestors = 218592628)

Zruty0 · 2018-09-19T00:10:58Z

src/Microsoft.ML.Data/Prediction/IPredictionTransformer.cs

+        TModel Model { get; }
+    }
+
+    public interface IClassicPredictionTransformer<out TModel> : IPredictionTransformer<TModel>


Classic [](start = 22, length = 7)

'Classic' sounds a bit wacky, even though it was my suggestion. Maybe 'SingleInput' ?.. #Resolved

Classic is weird.

In reply to: 218633462 [](ancestors = 218633462)

Zruty0 · 2018-09-19T00:13:38Z

src/Microsoft.ML.Data/Scorers/PredictionTransformer.cs


+        protected void SaveModel(ModelSaveContext ctx)
+        {
+            // *** Binary format ***


*** Binary format *** [](start = 15, length = 21)

whenever you save or load, need *** Binary format *** #Pending

?

In reply to: 218633828 [](ancestors = 218633828)

Zruty0 · 2018-09-19T00:14:23Z

src/Microsoft.ML.Data/Training/ITrainerEstimator.cs

@@ -7,7 +7,7 @@
 namespace Microsoft.ML.Runtime.Training
 {
    public interface ITrainerEstimator<out TTransformer, out TPredictor>: IEstimator<TTransformer>
-        where TTransformer: IPredictionTransformer<TPredictor>
+        where TTransformer: IClassicPredictionTransformer<TPredictor>


IClassicPredictionTransformer [](start = 28, length = 29)

I believe this change is incorrect, isn't it? #Pending

No, maybe i named in reverse, but this is the old interface.

In reply to: 218633926 [](ancestors = 218633926)

Zruty0 · 2018-09-19T00:17:35Z

...icrosoft.ML.StandardLearners/FactorizationMachine/FieldAwareFactorizationMachinePredictor.cs

+        public string[] FeatureColumns { get; }
+
+        /// <summary>
+        /// The type of the prediction transformer


The type of the prediction transformer [](start = 12, length = 38)

fix the comment #Resolved

Zruty0 · 2018-09-19T00:18:50Z

...icrosoft.ML.StandardLearners/FactorizationMachine/FieldAwareFactorizationMachinePredictor.cs

+                loaderSignature: LoaderSignature);
+        }
+
+        private static FieldAwareFactorizationMachinePredictionTransformer Create(IHostEnvironment env, ModelLoadContext ctx)


private [](start = 8, length = 7)

this is just to avoid having a public ctor? #Pending

Didn't load through the ctor; i think bc for the ctor the loadable class signature requires the args.

In reply to: 218634595 [](ancestors = 218634595)

Zruty0

TomFinley · 2018-09-19T14:08:55Z

src/Microsoft.ML.Data/Scorers/PredictionTransformer.cs

+        protected void SaveModel(ModelSaveContext ctx)
+        {
+            // *** Binary format ***
+            // model: prediction model.


model: prediction model. [](start = 15, length = 24)

Technically the model isn't part of this format, since you're not writing it to the stream, you're writing it somewhere else entirely, but that's OK. Consider fixing if you have to change the code anyway. #Resolved

@tfinley@gmail.com fixing it == remove the comment?

In reply to: 218818528 [](ancestors = 218818528)

TomFinley · 2018-09-19T14:11:07Z

src/Microsoft.ML.Data/Scorers/PredictionTransformer.cs

+        }
+    }
+
+        public abstract class ClassicPredictionTransformerBase<TModel> : PredictionTransformerBase<TModel>, IClassicPredictionTransformer<TModel>, ICanSaveModel


I guess by "classic" this just means a prediction transformer base that takes a single features column as its input. Classic is a bit of a funny word, but then again SinlgeFeaturesPredictionTransformerBase might be a bit of a mouthful and itself potentially confusing. #Resolved

TomFinley · 2018-09-19T14:13:16Z

src/Microsoft.ML.StandardLearners/Standard/MultiClass/Pkpd.cs

@@ -26,7 +26,7 @@ namespace Microsoft.ML.Runtime.Learners
 {

    using TDistPredictor = IDistPredictorProducing<float, float>;
-    using TScalarTrainer = ITrainerEstimator<IPredictionTransformer<IPredictorProducing<float>>, IPredictorProducing<float>>;
+    using TScalarTrainer = ITrainerEstimator<IClassicPredictionTransformer<IPredictorProducing<float>>, IPredictorProducing<float>>;


Classic [](start = 46, length = 7)

Yes after seeing this a bit I think the word "classic" is just going to confuse the heck out of a lot of people. Please consider doing something else. #Resolved

TomFinley · 2018-09-19T14:14:57Z

src/Microsoft.ML.Data/Prediction/IPredictionTransformer.cs

+        TModel Model { get; }
+    }
+
+    public interface IClassicPredictionTransformer<out TModel> : IPredictionTransformer<TModel>


IClassicPredictionTransformer [](start = 21, length = 29)

I sort of feel like an interface called IClassicPredictionTransformer needs some XML comment on it. All public interfaces should, but especially one with the word "classic" in the name.

I am choosing to interpret this interface as it gives me classic coke whenever I use it. #Resolved

TomFinley · 2018-09-19T14:15:36Z

test/Microsoft.ML.TestFramework/DataPipe/TestDataPipeBase.cs

-                CheckSameSchemas(scoredTrain.Schema, scoredTrain2.Schema);
-                CheckSameValues(scoredTrain, scoredTrain2);
+                  CheckSameSchemas(scoredTrain.Schema, scoredTrain2.Schema);
+                  CheckSameValues(scoredTrain, scoredTrain2);
            };


Somehow these got indented two spaces for no obvious reason. #Resolved

…on. Commenting code further.

TomFinley · 2018-09-19T18:00:36Z

src/Microsoft.ML.Data/Scorers/PredictionTransformer.cs

+    /// Base class for transformers with no feature column, or more than one feature columns.
+    /// </summary>
+    /// <typeparam name="TModel"></typeparam>
+    public abstract class PredictionTransformerBase<TModel> : IPredictionTransformer<TModel>


No documentation? #Pending

it is there? Are you looking at iteration 10?

In reply to: 218905403 [](ancestors = 218905403)

TomFinley

Looks pretty good thanks @sfilipi

sfilipi · 2018-09-20T19:13:03Z

close-reopen to trigger build

FAFM to extend TrainerEstimatorBase

b53d09e

sfilipi self-assigned this Sep 14, 2018

sfilipi requested review from Ivanidzo4ka, TomFinley, artidoro and Zruty0 September 14, 2018 06:39

sfilipi added the API Issues pertaining the friendly API label Sep 14, 2018

sfilipi added this to the 0918 milestone Sep 14, 2018

Zruty0 mentioned this pull request Sep 14, 2018

New API for ML.NET #754

Closed

sfilipi added 4 commits September 14, 2018 17:17

Merge branch 'master' into ffmEstimator

d7c942d

support for an array of features

6358777

FAFM doesn't inherit from TrainerEstimatorBase, just implements ITrainerEstimator

Merge branch 'master' into ffmEstimator

0e31686

incorporating metadata in the columns

67f41a3

sfilipi changed the title ~~[WIP] FAFM to extend TrainerEstimatorBase~~ FAFM to estimator Sep 18, 2018

sfilipi requested a review from wschin September 18, 2018 00:45

checkign whether any new regressions due to the work.

d4f5413

Zruty0 reviewed Sep 18, 2018

View reviewed changes

Zruty0 suggested changes Sep 18, 2018

View reviewed changes

reverting the change on IPredictionTransformer taking an array of fea…

26691e3

…tures. Splitting IPredictionTransformer into two interfaces Creating a transformer wrapping the FAFM predictor.

sfilipi commented Sep 18, 2018

View reviewed changes

Zruty0 reviewed Sep 19, 2018

View reviewed changes

Zruty0 approved these changes Sep 19, 2018

View reviewed changes

TomFinley reviewed Sep 19, 2018

View reviewed changes

sfilipi added 3 commits September 19, 2018 08:51

renaming the interface for transformers with feature column informati…

5890b11

…on. Commenting code further.

Fixing thy space.

174e75d

Incorporating Pete's fix about the unecessary creation of getters

65a6296

TomFinley reviewed Sep 19, 2018

View reviewed changes

TomFinley approved these changes Sep 19, 2018

View reviewed changes

sfilipi added 3 commits September 19, 2018 12:38

fixing comments.

492a890

merging from master

e86cfb9

post merge fixes

3d4858d

sfilipi closed this Sep 20, 2018

sfilipi reopened this Sep 20, 2018

sfilipi merged commit 044a6d3 into dotnet:master Sep 20, 2018

sfilipi deleted the ffmEstimator branch September 20, 2018 19:39

wschin changed the title ~~FAFM to estimator~~ Field-aware factorization machine to estimator Oct 15, 2018

ghost locked as resolved and limited conversation to collaborators Mar 29, 2022

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Field-aware factorization machine to estimator #912

Field-aware factorization machine to estimator #912

sfilipi commented Sep 14, 2018 •

edited

Loading

Zruty0 Sep 18, 2018 •

edited

Loading

sfilipi Sep 18, 2018 •

edited

Loading

Zruty0 Sep 18, 2018 •

edited by sfilipi

Loading

Zruty0 Sep 18, 2018 •

edited

Loading

Zruty0 Sep 18, 2018 •

edited

Loading

sfilipi Sep 18, 2018 •

edited

Loading

Zruty0 left a comment

sfilipi Sep 18, 2018 •

edited

Loading

sfilipi Sep 19, 2018

Zruty0 Sep 19, 2018 •

edited by sfilipi

Loading

TomFinley Sep 19, 2018

Zruty0 Sep 19, 2018 •

edited by sfilipi

Loading

sfilipi Sep 19, 2018

Zruty0 Sep 19, 2018 •

edited by sfilipi

Loading

sfilipi Sep 19, 2018

Zruty0 Sep 19, 2018 •

edited by sfilipi

Loading

Zruty0 Sep 19, 2018 •

edited by sfilipi

Loading

sfilipi Sep 19, 2018

Zruty0 left a comment

TomFinley Sep 19, 2018 •

edited by sfilipi

Loading

sfilipi Sep 19, 2018 •

edited

Loading

TomFinley Sep 19, 2018 •

edited by sfilipi

Loading

TomFinley Sep 19, 2018 •

edited by sfilipi

Loading

TomFinley Sep 19, 2018 •

edited by sfilipi

Loading

TomFinley Sep 19, 2018 •

edited by sfilipi

Loading

TomFinley Sep 19, 2018 •

edited by sfilipi

Loading

sfilipi Sep 19, 2018

TomFinley left a comment

sfilipi commented Sep 20, 2018


		IEstimator<ITransformer> est = new FieldAwareFactorizationMachineTrainer(Env, "Label", new[] { "Feature1", "Feature2", "Feature3", "Feature4" });

		//var result = est.Fit(data);

Field-aware factorization machine to estimator #912

Field-aware factorization machine to estimator #912

Conversation

sfilipi commented Sep 14, 2018 • edited Loading

Zruty0 Sep 18, 2018 • edited Loading

Choose a reason for hiding this comment

sfilipi Sep 18, 2018 • edited Loading

Choose a reason for hiding this comment

Zruty0 Sep 18, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

Zruty0 Sep 18, 2018 • edited Loading

Choose a reason for hiding this comment

Zruty0 Sep 18, 2018 • edited Loading

Choose a reason for hiding this comment

sfilipi Sep 18, 2018 • edited Loading

Choose a reason for hiding this comment

Zruty0 left a comment

Choose a reason for hiding this comment

sfilipi Sep 18, 2018 • edited Loading

Choose a reason for hiding this comment

sfilipi Sep 19, 2018

Choose a reason for hiding this comment

Zruty0 Sep 19, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

TomFinley Sep 19, 2018

Choose a reason for hiding this comment

Zruty0 Sep 19, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

sfilipi Sep 19, 2018

Choose a reason for hiding this comment

Zruty0 Sep 19, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

sfilipi Sep 19, 2018

Choose a reason for hiding this comment

Zruty0 Sep 19, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

Zruty0 Sep 19, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

sfilipi Sep 19, 2018

Choose a reason for hiding this comment

Zruty0 left a comment

Choose a reason for hiding this comment

TomFinley Sep 19, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

sfilipi Sep 19, 2018 • edited Loading

Choose a reason for hiding this comment

TomFinley Sep 19, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

TomFinley Sep 19, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

TomFinley Sep 19, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

TomFinley Sep 19, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

TomFinley Sep 19, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

sfilipi Sep 19, 2018

Choose a reason for hiding this comment

TomFinley left a comment

Choose a reason for hiding this comment

sfilipi commented Sep 20, 2018

sfilipi commented Sep 14, 2018 •

edited

Loading

Zruty0 Sep 18, 2018 •

edited

Loading

sfilipi Sep 18, 2018 •

edited

Loading

Zruty0 Sep 18, 2018 •

edited by sfilipi

Loading

Zruty0 Sep 18, 2018 •

edited

Loading

Zruty0 Sep 18, 2018 •

edited

Loading

sfilipi Sep 18, 2018 •

edited

Loading

sfilipi Sep 18, 2018 •

edited

Loading

Zruty0 Sep 19, 2018 •

edited by sfilipi

Loading

Zruty0 Sep 19, 2018 •

edited by sfilipi

Loading

Zruty0 Sep 19, 2018 •

edited by sfilipi

Loading

Zruty0 Sep 19, 2018 •

edited by sfilipi

Loading

Zruty0 Sep 19, 2018 •

edited by sfilipi

Loading

TomFinley Sep 19, 2018 •

edited by sfilipi

Loading

sfilipi Sep 19, 2018 •

edited

Loading

TomFinley Sep 19, 2018 •

edited by sfilipi

Loading

TomFinley Sep 19, 2018 •

edited by sfilipi

Loading

TomFinley Sep 19, 2018 •

edited by sfilipi

Loading

TomFinley Sep 19, 2018 •

edited by sfilipi

Loading

TomFinley Sep 19, 2018 •

edited by sfilipi

Loading