SelectFeaturesBasedOn* Transformers are not outputing vectors of known size.

[nsingal]

System information

• OS version/distro: Windows
• .NET Version (eg., dotnet --info):

Code that failed:

Context.Transforms.Text.TokenizeIntoWords("TextFeature") .Append(Context.Transforms.FeatureSelection.SelectFeaturesBasedOnCount("TextFeature", "TextFeature", 10)).
Append(Context.Transforms.FeatureSelection.SelectFeaturesBasedOnMutualInformation("TextFeature", "TextFeature", "Label", 5000))
.Append(Context.Transforms.NormalizeLpNorm("TextFeature", "TextFeature", Microsoft.ML.Transforms.LpNormNormalizingEstimatorBase.NormFunction.L2, true));

Error:
At NormalizeLpNorm, it fails with
Schema mismatch for input column 'TextFeature': expected Expected Single or known-size vector of Single, got VarVector

[yaeldekel]

After investigating, I discovered that the bug is actually in the WordBagEstimator. In PR #4702 I am adding early checks for the input types in the feature selection estimators.

[mstfbl]

This issue is fixed by #4702 .