TrainTestSplit function #1005
TrainTestSplit function #1005
Conversation
| /// train to the test set.</remarks> | ||
| /// <returns>A pair of datasets, for the train and test set.</returns> | ||
| public static (DataView<T> trainSet, DataView<T> testSet) TrainTestSplit<T>(this TrainContextBase context, | ||
| DataView<T> data, double testFraction = 0.1, Func<T, PipelineColumn> stratificationColumn = null) |
There was a problem hiding this comment.
Func<T, PipelineColumn> [](start = 57, length = 23)
Are there any type restrictions on this? It must be hashable, I suppose? #Resolved
There was a problem hiding this comment.
yea, it must be hash-able, but we should bring back the hash-join functionality for all the column types. I will write an issue about this
In reply to: 220069503 [](ancestors = 220069503)
There was a problem hiding this comment.
I have a REVIEW comment about it in the TrainContext, line 140
In reply to: 220366016 [](ancestors = 220366016,220069503)
| @@ -236,7 +236,7 @@ public static (Vector<float> score, Key<uint, TVal> predictedLabel) | |||
| { | |||
| Contracts.CheckValue(label, nameof(label)); | |||
| Contracts.CheckValue(features, nameof(features)); | |||
| Contracts.CheckValue(loss, nameof(loss)); | |||
| Contracts.CheckValueOrNull(loss); | |||
There was a problem hiding this comment.
Err, whoops, thanks for fixing this. 👍
There was a problem hiding this comment.
This looks mostly good. I think we ought to think carefully about what we feel about having the training context responsible for doing things like splitting data, but I feel this might be best done by merely observing the idioms you've proposed in action. Thanks @Zruty0 .
| r.features, | ||
| maxIterations: 2))); | ||
|
|
||
| var results = ctx.CrossValidate(reader.Read(dataSource), est, r => r.label) |
There was a problem hiding this comment.
I like how the example/test flows. #ByDesign
| for (int fold = 0; fold < numFolds; fold++) | ||
| result.Add(foldFunction(fold)); | ||
|
|
||
| return result.ToArray(); |
There was a problem hiding this comment.
result.ToArray() [](start = 19, length = 16)
curious, why not return a list? #Resolved
There was a problem hiding this comment.
|
|
||
| if (stratificationColumn == null) | ||
| { | ||
| stratificationColumn = data.Schema.GetTempColumnName("StratificationColumn"); |
There was a problem hiding this comment.
StratificationColumn [](start = 70, length = 20)
is there a DefaultColumnName for this? #Resolved
There was a problem hiding this comment.
I looked, there is none. I don't think there should be either, it's a pretty peculiar concept, unique to CV and split scenarios
In reply to: 220372046 [](ancestors = 220372046)
ghost
locked as resolved and limited conversation to collaborators
Part of #754
Added TrainTestSplit function to training contexts.