Dependency Error when using Ordinary Least Squares Regressor

[jwood803]

System information

Product Information:
 Version:            2.1.201
 Commit SHA-1 hash:  7932dc6179

Runtime Environment:
 OS Name:     Windows
 OS Version:  10.0.17134
 OS Platform: Windows
 RID:         win10-x64
 Base Path:   C:\Program Files\dotnet\sdk\2.1.201\

Microsoft .NET Core Shared Framework Host

  Version  : 2.0.7
  Build    : 2d61d0b043915bc948ebf98836fefe9ba942be11

Issue

• What did you do?
  Created a simple ML.NET Console project with the following code:

  var pipeline = new LearningPipeline();
  
  var splitter = new TrainTestDatasetSplitter() { Fraction = 0.7f };
  
  pipeline.Add(new TextLoader<SalaryData>("SalaryData.csv", useHeader: true, separator: ","));
  pipeline.Add(new ColumnConcatenator("Features", "YearsExperience"));
  pipeline.Add(new OrdinaryLeastSquaresRegressor());
  
  pipeline.Train<SalaryData, SalaryPrediction>();
  

• What happened?
  When calling Train, I'm getting the error The MKL library (Microsoft.ML.MklImports.dll) or one of its dependencies is missing. After some searching, I couldn't find where I can get a reference to it to include in the project.

• What did you expect?
  Expected no errors, or a way to include the missing dependency.

If anything else is needed, just let me know. Thanks!

[Ivanidzo4ka]

machinelearning/src/Microsoft.ML.StandardLearners/Standard/OlsLinearRegression.cs

Line 360 in 5d368b8

internal static class Mkl

This is dependency on MKL library which we use internally, but currently it's unclear how to consume it because of legal and infrastructure issues.

Current options are:

1. find way to add MKL as dependency
2. rewrite LAPACKE_dpptrf LAPACKE_dpptrs LAPACKE_dpptri in CpuMathNative
3. disable learner (on entry point level or just remove it from repo)

@GalOshri @glebuk @shauheen