Adding a Shuffler Transformer

[AtheMathmo]

Note that this is blocked by #117 .

With the addition of swap_rows and swap_cols it would be nice if we added a Transformer to shuffle the rows of input data. We should use an algorithm similar to the in_place_fisher_yates in learning/toolkit/rand_utils to do the actual shuffle.

There are a couple ways that we can handle the inv_transform function.

• Don't implement it at all and just return an Err, this is really ugly.
• Implement it by keeping track of the swaps and doing them in reverse order. Has some memory overhead for a feature that will rarely be needed.
• Separate out the traits: pub trait Invertible: Transformer. And don't implement Invertible for Shuffler.

None of these solutions feel all that clean. We could adopt the same approach as DBSCAN and hide the memory overhead behind a flag on the Shuffler struct. This is really ugly but does allow the user to pick the best option for them.

Would be happy to get feedback on these options.

[AtheMathmo]

There is a POC on branch shuffler - it needs the rulinalg 0.3 update so that we can use swap_rows.