Support dynamic types when working with IDataView

[baruchiro]

Issue

I want to read my data from the .csv file. I think it may be done without strongly-typing my objects. In C#, we have two potential abilities: Anonymous Types and Dynamic Types.

I tried to check how the existing code could be adapted, but I could not fully understand the code.

[baruchiro]

Same in MLContext.Data.CreateEnumerable<>. If we've transformed our DataView, how can we get objects from it?

Cross-linking to SO question.

[yaeldekel]

Hi @baruchiro,
Currently in order to load data from a text file, you need to provide a schema, either by specifying the column definitions, like here and here, or by specifying a type, with public fields or properties that have LoadColumnAttributes to help translate these fields and properties into physical columns in the text file.

The AutoML branch has experimental column inferencing functionality that tries figuring out the columns and their types from looking at the dataset.

Regarding your second question, CreateEnumerable is an API for getting objects from an IDataView. The type of the objects has to contain settable fields or properties with types that match the column types in the IDataView schema. Here is an example of how to use this API.
Please let me know whether this answers your question.

[baruchiro]

This is more a feature request than a question.

[sandorfr]

Here is an example of how to use this API.
Please let me know whether this answers your question.

From the look of it this example is dynamic only by the name of it. It's using a static strongly typed POCO.

At the moment it seems that the only way to do handle dynamic data structure for would be to generate and build code.