[Data Tests] Add ability to test your data

[aslotte]

Is your feature request related to a problem? Please describe.
Part of the model flow is to create and add data validation tests. These may check the sanity of your data, e.g. that certain columns have a specific cardinality (e.g. only 5 different kinds of values), or that a numeric data column has a specific range.

Creating these tests are pretty repetitive.

Describe the solution you'd like
Add a Fluent API to validate data structure.

I'm envisioning a syntax such as the following in a new project, e.g. MLOps.NET.Data.Tests

[TestMethod]
public void VerifyCardinalityOfColumn()
{
    var mlOpsTestingContext = new MLOpsTestingContext();

    mlOpsTestingContext.WithData(pathToData) 
        .HasColumn(index, x => x.WithCardinality(3))
        .Assert()
}

[TestMethod]
public void VerifyRangeOfColumn()
{
    var mlOpsTestingContext = new MLOpsTestingContext();

    mlOpsTestingContext.WithData(pathToData) 
        .HasColumn(index, x => x.WithRange(min: 0, max: 10000)
        .Assert()
}

[TestMethod]
public void VerifySchema()
{
    var mlOpsTestingContext = new MLOpsTestingContext();

    mlOpsTestingContext.WithData(pathToData) 
        .HasNumberOfColumns(10)
        .HasMinimumNumberOfRows(5000)
        .Assert()
}

[TestMethod]
public void VerifyColumOnlyContainsApprovedValues()
{
    var mlOpsTestingContext = new MLOpsTestingContext();

    mlOpsTestingContext.WithData(pathToData) 
        .HasColumn(index, x => x.WithValues(listOfApprovedValues)
        .Assert()
}

[aslotte]

@lqdev working on the workshop material, I thought that something like this would be super useful to create data validation tests as part of the pipeline.

[aslotte]

We can probably implement some of these checks using the new DataFrame API (at least it would be fun to try)

[AnoojNair]

@aslotte , Nice,That would be a cool feature to have w.r.t a ML workflow. Would it be part of MLopsContext or a separate context?

[aslotte]

It would be a separate context I think, as it wouldn't need the same dependencies we currently use.