Xshard max, split, and select on Ray

[jenniew]

Add min/max operation. Related issue: https://github.com/analytics-zoo/orca/issues/6
Add train_test_split operation. Related issue: https://github.com/analytics-zoo/orca/issues/4
Add selection with column name/slice range. Related issue: https://github.com/analytics-zoo/orca/issues/5
Add **kwargs support for read_csv, read_json. Related issue:https://github.com/analytics-zoo/orca/issues/20
Change DataShards.apply to DataShards.transform_shard.
Add unit tests for all new APIs.
Fix xshard spark issues.
Refactor code.

[jason-dai]

test_transform_shard

[jason-dai]

change to transform

[jason-dai]

change to transform

[jason-dai]

There are too many changes in this PR, which makes it difficult to review and verify. We should make sure each PR only contains the minimum feature sets. One fundamental issue with the changes is that how to maintain multiple DataShards with one common set of underlying Actors. For instance,

s = xshard.pands.read_csv(...)
s1 = s['user_id']
s1.tranform_shard(func) #what is the expected input of func? will it also change s?

And

s = xshard.pands.read_csv(...)
s1, s2 = train_test_split(s)
s.transform_shard(func) #will it change s1 and s2?

I suggest we close this PR and open several smaller ones (e.g., fixing https://github.com/analytics-zoo/orca/issues/20 first)

[jenniew]

There are too many changes in this PR, which makes it difficult to review and verify. We should make sure each PR only contains the minimum feature sets. One fundamental issue with the changes is that how to maintain multiple DataShards with one common set of underlying Actors. For instance,

s = xshard.pands.read_csv(...)
s1 = s['user_id']
s1.tranform_shard(func) #what is the expected input of func? will it also change s?

And

s = xshard.pands.read_csv(...)
s1, s2 = train_test_split(s)
s.transform_shard(func) #will it change s1 and s2?

I suggest we close this PR and open several smaller ones (e.g., fixing analytics-zoo/orca#20 first)

The current transform would change data in s. We need to change the implementation to return new DataFrame or Series. Change Actor implementation or just use object id no Actor.
Already create PR: https://github.com/intel-analytics/analytics-zoo/pull/2305 for (https://github.com/analytics-zoo/orca/issues/20).
Close this PR.