Add samples in TT for FFM #3312
Conversation
| @@ -0,0 +1,74 @@ | |||
| using System; | |||
There was a problem hiding this comment.
This file is produced by TT. Please ignore the difference here because it replaces the old file entirely.
| @@ -0,0 +1,84 @@ | |||
| using System; | |||
There was a problem hiding this comment.
This file is produced by TT. Please ignore the difference here because it replaces the old file entirely.
| @@ -1,84 +1,167 @@ | |||
| using System; | |||
| using System.Collections.Generic; | |||
There was a problem hiding this comment.
This file is produced by TT. Please ignore the difference here because it replaces the old file entirely.
| @@ -1,76 +1,157 @@ | |||
| using System; | |||
| using System.Collections.Generic; | |||
There was a problem hiding this comment.
This file is produced by TT. Please ignore the difference here because it replaces the old file entirely.
#Resolved
Codecov Report
@@ Coverage Diff @@
## master #3312 +/- ##
==========================================
- Coverage 72.65% 72.64% -0.01%
==========================================
Files 807 807
Lines 145191 145191
Branches 16223 16223
==========================================
- Hits 105485 105481 -4
- Misses 35290 35293 +3
- Partials 4416 4417 +1
|
Use proper namespace Remove old samples Fix typo Fix csproj
| var metrics = mlContext.BinaryClassification.Evaluate(transformedTrainingData); | ||
|
|
||
| // Show the quality metrics. | ||
| Microsoft.ML.SamplesUtils.ConsoleUtils.PrintMetrics(metrics); |
There was a problem hiding this comment.
Microsoft.ML.SamplesUtils [](start = 12, length = 25)
we're deprecating SamplesUtils. Let's add the helper to the .ttincludefile #Resolved
|
|
||
| // Expected output: | ||
| // Accuracy: 0.99 | ||
| // AUC: 1.00 |
There was a problem hiding this comment.
1.00 [](start = 22, length = 4)
[optional] can we make it a little harder for the classifier? maybe reduce the label delta from 0.2f to 0.05f or some lower number? #ByDesign
There was a problem hiding this comment.
Maybe no? It's super important to show a case where a model works well. This is a motive way to push user to think about feature engineering if my model doesn't work on their case. #Resolved
| } | ||
| } | ||
| } | ||
|
|
| <#+ | ||
| string ClassName="FieldAwareFactorizationMachine"; | ||
| string Trainer = @"FieldAwareFactorizationMachine( | ||
| // Specify three feature columns! |
There was a problem hiding this comment.
// Specify three feature columns! [](start = 16, length = 33)
Majestic.
Maybe it's worth to mention what most of the trainers support one feature column, and why person should pass three separate feature columns instead of concat them into one.
Or maybe it's should be part of xml documentation and not related to you PR, what do I know about documentation anyway. #Resolved
There was a problem hiding this comment.
There is no ttinclude in Sample's csproj. In reply to: 483329386 [](ancestors = 483329386) Refers to: docs/samples/Microsoft.ML.Samples/Dynamic/Trainers/BinaryClassification/MultipleFeatureColumnsBinaryClassification.ttinclude:1 in 4b8cd50. [](commit_id = 4b8cd50, deletion_comment = False) |
ghost
locked as resolved and limited conversation to collaborators
3 participants
Two samples in for FFM are added following new sample guideline. Related to #2522.