You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
There was a prior PR #2581 and issue #2501 related to value-tuples and why they should not be part of our public surface. I have noticed that there are some "stragglers" still remaining in the public API. So, the work is perhaps not yet complete.
The following list is I believe complete for Core/Data/Transforms/FastTree/ImageAnalytics/KMeansClustering/LightGBM/PCA/Tensorflow/StandardLearners/Data.DataView assemblies.
There are three distinct categories where this flaw has remained. (Though the last "category" other has only one item.)
Properties on transformers
Some transformers are exposing information about themselves via this mechanism.
KeyToBinaryVectorMappingTransformer.Columns
MissingValueDroppingTransformer.Columns
MissingValueIndicatorTransformer.Columns
CustomStopWordsRemovingTransformer.Columns
TextNormalizingTransformer.Columns
TokenizingByCharactersTransformer.Columns
WordEmbeddingsExtractingTransformer.Columns
ImageGrayscalingTransformer.Columns
ImageLoadingTransformer.Columns
LatentDirichletAllocationTransformer.ItemScoresPerTopic and WordScoresPerTopic
MLContext estimator creation extension methods
There are some overloads of MLContext extension methods on various catalogs that are stil using it. I view this as a lesser sin since this is at least something that could conceivably be fixed using an overload if we decide it is necessary, but I'd still prefer to be consistent.
ProduceHashedNgrams extension method
ProduceHashedWordBags extension method
ProduceNgrams extension method
ProduceWordBags extension method
RemoveDefaultStopWords extension method
TokenizeWords extension method
Others
Lastly, I see a Microsoft.ML.ColumnOptions global class with an implicit operator from value-tuples. This one is probably harmless, since that specific class is for representing a simple case.
Well eventually, yes. If you want to go and sweep them in one PR, I wouldn't stop you. You will make life of scrubbing people easier.
But ask yourself a question, do you really want to make someone else life easier? That's not the Tom I used to knew! :D
There was a prior PR #2581 and issue #2501 related to value-tuples and why they should not be part of our public surface. I have noticed that there are some "stragglers" still remaining in the public API. So, the work is perhaps not yet complete.
The following list is I believe complete for
Core
/Data
/Transforms
/FastTree
/ImageAnalytics
/KMeansClustering
/LightGBM
/PCA
/Tensorflow
/StandardLearners
/Data.DataView
assemblies.There are three distinct categories where this flaw has remained. (Though the last "category" other has only one item.)
Properties on transformers
Some transformers are exposing information about themselves via this mechanism.
MLContext estimator creation extension methods
There are some overloads of MLContext extension methods on various catalogs that are stil using it. I view this as a lesser sin since this is at least something that could conceivably be fixed using an overload if we decide it is necessary, but I'd still prefer to be consistent.
Others
Lastly, I see a Microsoft.ML.ColumnOptions global class with an implicit operator from value-tuples. This one is probably harmless, since that specific class is for representing a simple case.
/cc @yaeldekel @Ivanidzo4ka
The text was updated successfully, but these errors were encountered: