TensorFlowTransformer is not properly matching the input dimensions

[LucasFievet]

System information

• OS version/distro: Windows 10/Pro
• .NET Version (eg., dotnet --info): .NET Version 2.1.202
• ML.TensorFlow Version 0.10

Issue

• What did you do?: Executed a TensorFlow model using TensorFlowEstimator
• What happened?: Mapper does not properly match the input dimensions to the TensorFlow model dimensions
• What did you expect?: Properly matched dimensions

Source code / logs

TensorFlowTransform.cs, Lines 810 - 875

Line 853:

var d = originalShape.NumDimensions > 2 ? Math.Pow(typeValueCount / valCount, 1.0 / (originalShape.NumDimensions - 2)) : 1;

The case with 2 dimensions, where the originalShape is [1, ?] is not matched properly.

[Ivanidzo4ka]

@yaeldekel

machinelearning/src/Microsoft.ML.TensorFlow/TensorflowTransform.cs

Line 853 in 453f191

var d = originalShape.NumDimensions > 2 ? Math.Pow(typeValueCount / valCount, 1.0 / (originalShape.NumDimensions - 2)) : 1;

do you know why we setting variable length only for tensors with size more than 2?
@zeahmed do we have tests with variable length? If yes, I believe it's probably for images, can we have test for two dimensional model?

[zeahmed]

No, this is the bug and I am going to make a fix for it.