TrainTestSplit should be inside MLContext.Data

[artidoro]

Currently TrainTestSplit can be found in the various training tasks in MLContext:

machinelearning/test/Microsoft.ML.Tests/Scenarios/Api/CookbookSamples/CookbookSamples.cs

Lines 604 to 605 in 533171a

	// Split the data 90:10 into train and test sets, train and evaluate.
	var (trainData, testData) = mlContext.MulticlassClassification.TrainTestSplit(data, testFraction: 0.1);

Since that is a data operation that only involves splitting data and since it is independent on the training task, it would make more sense to have it in mlContext.Data. The previous line would look like:

var (trainData, testData) = mlContext.Data.TrainTestSplit(data, testFraction: 0.1);

[TomFinley]

since it is independent on the training task,

Well, not exactly. In the case of ranking, for example, it is critical that the split be stratified on a sort of group ID. But we might consider that we expect that having the stratification is an option, and if someone doesn't realize stratification is important for a task like ranking, well... that's just too bad.

But I would at least be aware of this issue.

Also this does imply a breaking change, so I'm adding it to Project 13.