[dcostea]

I added WriteCsv method (plus some unit tests) to write a DataFrame to CSV file. This solves #2917.

@eerhardt @pgovind I have recreated the PR following Eric's recommendations from #2945. Please review.

[pgovind]

Apologies for the late review. I left some comments now :)

[dcostea]

Apologies for the late review. I left some comments now :)

Do not worry. I'm sure you guys are very busy with .NET 5 :)

[eerhardt]

(minor perf question/suggestion): I wonder if we used the IDataView API here we could avoid the boxing and unboxing of every primitive value in the DataFrame.

For example, for each column, we can get a Getter and use it to get the value generically and write it to the StringBuilder.

See https://github.com/dotnet/machinelearning/blob/5370692f6d1e6f1f686596d530032b4226dc9e80/src/Microsoft.ML.Data/DataDebuggerPreview.cs#L44-L60 for a general structure on how to do this.

I think what we have here is good for a first implementation to get the functionality added. But we may want to benchmark alternatives to see how well this performs.

[eerhardt]

I think this is pretty close. After a couple minor nits I think we can merge this.

[pgovind]

Just the 1 comment to rename a var. Everything else looks really good! Great work!

[dcostea]

Just the 1 comment to rename a var. Everything else looks really good! Great work!

Thank you everyone for all the help! :)