Proposal: ReadCsv schema provider as lambda function

[pkese]

The combination of inferTypes and schema is rather unflexible.

In my case (I'm trying to process lots of slightly different .csv files) I'd much prefer if I was able to provide types based on actual column titles that the reader found in the .csv file.

I'd propose a typeResolver parameter to ReadCsv that would accept a lambda function that resolves column titles to target types:

    Frame.ReadCsv("sample.csv", typeResolver=(fun title ->
        match title with
        | t when t.contains "Name" -> Some "string"
        | t when t.endsWith "Id" -> Some "guid"
        | t when t.isCapitalized -> Some "float?"
        | _ -> None))

My aim at this point is to:

• discuss this proposal
• discuss the API (e.g. should it be a typeResolver of type column: string -> type: string option or an even more flexible schemaGenerator of type columns: string list -> schema: string)
• figure out if it is OK to discuss this in the scope of Deedle or would CsvProvider project be the more appropriate place

I'm willing to look into the issue and try to provide an implementation as well.

[tpetricek]

This would not work for CSV provider, because CSV provider needs to take a literal as a parameter (i.e. just a string) but it sounds pretty sensible for CSV parser in Deedle.

I'd be happy with this but perhaps also:

• should the function take just a title, or should it also take the inferred type, so that it can use that if it wants to?
• should this return type as a string, or would it make sense to return a System.Type (and perhaps a bool to indicate whether the column can have missing values)?