[C++] CSV reader: Ability to not infer column types.

[asfimport]

I'm trying to read CSV as is. All columns as strings. I don't know the schema of these CSVs and they will vary as they are provided by user.

Right now i'm using pandas.read_csv(dtype=str) which works great, but since final destination of these CSVs are parquet files it seems like much more efficient to use pyarrow.csv.read_csv in future, as soon as this becomes available :)

I tried things like pyarrow.csv.read_csv(convert_types=ConvertOptions(columns_types=defaultdict(lambda: 'string'))) but it doesn't work.

Maybe I just didnt' find something that already exists? :)

Environment: Ubuntu Xenial
Reporter: Bogdan Klichuk

_{Note: This issue was originally created as ARROW-5811. Please see the migration documentation for further details.}

[asfimport]

Antoine Pitrou / @pitrou:
No, convert_types must be the full mapping of column names to data types. C++ doesn't know about defaultdict...

We could add more inference options, though, for example to select the datatypes for which inference is enabled.

[asfimport]

Antoine Pitrou / @pitrou:
@wesm @nealrichardson do you have an idea about a desirable API here?

[asfimport]

Wes McKinney / @wesm:
I think we need to create an abstract C++ type (or similar) that is a ConversionRule. We have other types of conversion rules where we have not defined an API yet, for example "timestamp with striptime-like format of $FORMAT". Whatever API we have, it needs to be extensible to accommodate new kinds of logic

[asfimport]

Neal Richardson / @nealrichardson:
I think I'm not understanding the problem. What's missing from the column_types we already support? https://github.com/apache/arrow/blob/master/cpp/src/arrow/csv/options.h#L69

[asfimport]

Antoine Pitrou / @pitrou:
The request is for no inference to occur, without knowing the column names or the number of columns in advance (so you cannot pass an explicit column_types).

[asfimport]

Neal Richardson / @nealrichardson:
In principle, a user could parse the header row of the CSV separately to identify the column names, then use that to define column_types mapping each name to string type. So are we just talking about how to facilitate that, whether/how to internalize that logic and expose it as a simple argument? Or is there something else?

If column_types didn't have to be a map, maybe that would help. Perhaps it could also accept an array of length equal to the number of columns, or a single value, in which case it would recycle that type for every column. 

 

[asfimport]

Antoine Pitrou / @pitrou:
We're talking about C++ here. Dynamic typing isn't terribly idiomatic (though it's possible using std::variant) :-)

[asfimport]

Wes McKinney / @wesm:
Yeah, so we could define a conversion rule to return string or binary, and then add an option to set a default conversion rule (where currently we have an implicit default of "use type inference")