DnnFeaturizeImage sample doesn't match the signature in the current API

[yaeldekel]

The API for LoadImage is

LoadImages(string outputColumnName, string imageFolder, string inputColumnName = null)

but in the sample it is used like this:

mlContext.Transforms.LoadImages(imagesFolder, "ImageObject", "ImagePath")

One of them should be fixed, not sure which one? @TomFinley , @artidoro , @wschin
I kind of like the second option (the folder first, output and input column names together).

[rogancarr]

So, usually the Catalog Extension Methods have APIs like outputColumnName, inputColumnName, specific-arguments. However, there are some that have inputColumnName shoved to the back to make it nullable.

I'd like to uplevel this question to the two questions:

1. "Is it okay to shove inputColumnName to the back to make it nullable?
2. "Is it okay to shove inputColumnNames (arrays) to the back to allow them to be params?

Here are a list of APIs that do No. 1:

MapValue

ValueMappingEstimator<TInputType, TOutputType> MapValue<TInputType, TOutputType>(
            this TransformsCatalog.ConversionTransforms catalog,
            string outputColumnName,
            IEnumerable<KeyValuePair<TInputType, TOutputType>> keyValuePairs,
            string inputColumnName = null,
            bool treatValuesAsKeyType = false);

public static ValueMappingEstimator<TInputType, TOutputType> MapValue<TInputType, TOutputType>(
            this TransformsCatalog.ConversionTransforms catalog,
            string outputColumnName,
            IEnumerable<KeyValuePair<TInputType, TOutputType[]>> keyValuePairs,
            string inputColumnName = null);

        public static ValueMappingEstimator MapValue(
            this TransformsCatalog.ConversionTransforms catalog,
            string outputColumnName, IDataView lookupMap, DataViewSchema.Column keyColumn, DataViewSchema.Column valueColumn, string inputColumnName = null)

DNNFeaturizeImage

        public static DnnImageFeaturizerEstimator DnnFeaturizeImage(this TransformsCatalog catalog,
            string outputColumnName,
            Func<DnnImageFeaturizerInput, EstimatorChain<ColumnCopyingTransformer>> modelFactory,
            string inputColumnName = null);

ResizeImages

        public static ImageResizingEstimator ResizeImages(this TransformsCatalog catalog,
            string outputColumnName,
            int imageWidth,
            int imageHeight,
            string inputColumnName = null,
            ImageResizingEstimator.ResizingKind resizing = ImageResizingEstimator.ResizingKind.IsoCrop,
            ImageResizingEstimator.Anchor cropAnchor = ImageResizingEstimator.Anchor.Center);

ApplyWordEmbedding

        public static WordEmbeddingEstimator ApplyWordEmbedding(this TransformsCatalog.TextTransforms catalog,
            string outputColumnName,
            string customModelFile,
            string inputColumnName = null)

And here are the APIs that do No. 2:

TextFeaturizer

        public static TextFeaturizingEstimator FeaturizeText(this TransformsCatalog.TextTransforms catalog,
            string outputColumnName,
            TextFeaturizingEstimator.Options options,
            params string[] inputColumnNames);

And here is an API that just seems weird:

ApplyOnnxModel just has a funky order:

public static OnnxScoringEstimator ApplyOnnxModel(this TransformsCatalog catalog,
            string modelFile,
            string outputColumnName,
            string inputColumnName,
            int? gpuDeviceId = null,
            bool fallbackToCpu = false);

[rogancarr]

We spoke about this offline. Standards were set in #2064.

The first two sets of APIs are correct, and should not be changed. The LoadImages discrepancy can be fixed by updating the Sample.

The ApplyOnnxModel API is incorrect and has been posted in issue #3082.