html_url,issue_url,id,node_id,user,created_at,updated_at,author_association,body,reactions,performed_via_github_app,issue
https://github.com/pydata/xarray/pull/1289#issuecomment-290604141,https://api.github.com/repos/pydata/xarray/issues/1289,290604141,MDEyOklzc3VlQ29tbWVudDI5MDYwNDE0MQ==,1217238,2017-03-31T03:10:53Z,2017-03-31T03:10:53Z,MEMBER,This is great -- thanks @fujiisoup !,"{""total_count"": 1, ""+1"": 1, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,211323643
https://github.com/pydata/xarray/pull/1289#issuecomment-289036993,https://api.github.com/repos/pydata/xarray/issues/1289,289036993,MDEyOklzc3VlQ29tbWVudDI4OTAzNjk5Mw==,6815844,2017-03-24T14:26:16Z,2017-03-24T14:26:16Z,MEMBER,"@shoyer 
Thank you for the reply.
I did not yet think of smarter solution.
I hope the current implementation is acceptable. 
If I come up with cleaner script, I will send another PR.","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,211323643
https://github.com/pydata/xarray/pull/1289#issuecomment-288866331,https://api.github.com/repos/pydata/xarray/issues/1289,288866331,MDEyOklzc3VlQ29tbWVudDI4ODg2NjMzMQ==,5635139,2017-03-23T21:30:50Z,2017-03-23T21:30:50Z,MEMBER,"> I'll let @MaximilianR give the final review 

Apologies for missing this. I actually think @jhamman has done much more than I to contribute to this area!

But this change looks good, thanks @fujiisoup !","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,211323643
https://github.com/pydata/xarray/pull/1289#issuecomment-286665079,https://api.github.com/repos/pydata/xarray/issues/1289,286665079,MDEyOklzc3VlQ29tbWVudDI4NjY2NTA3OQ==,2443309,2017-03-15T07:45:51Z,2017-03-15T07:45:51Z,MEMBER,"@fujiisoup - This is looking pretty good. I'll let @MaximilianR give the final review since he's spent more time here than I have. As for your four points above, I think I'm fine with all of those design decisions.

One minor point though (no change needed)

> All the attrs are dropped

I don't think this is true for data_vars that don't have rolling-dim in them.","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,211323643
https://github.com/pydata/xarray/pull/1289#issuecomment-284367479,https://api.github.com/repos/pydata/xarray/issues/1289,284367479,MDEyOklzc3VlQ29tbWVudDI4NDM2NzQ3OQ==,6815844,2017-03-06T11:08:09Z,2017-03-06T11:08:09Z,MEMBER,">  we try to do it when possible because it leads to more predictable results.

OK. I totally agree with this.
(Today I experienced a bug caused by unexpectedly aligned data_vars in other project. I realized predictable behavior is really important :) )

I think this PR is ready for review. If anyone has comments about the following behavior, I'm happy to update the code.
+ Only data_vars depending on the given rolling-dim are reduced. The others remain unchanged but are still stored in the reduced dataset (not dropped).
+ If rolling-dim is not included in `Dataset.dims`, this class raises `KeyErrror`.
+ The order of data_vars remain unchanged even after `reduce()`.
+ All the attrs are dropped.","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,211323643
https://github.com/pydata/xarray/pull/1289#issuecomment-284278244,https://api.github.com/repos/pydata/xarray/issues/1289,284278244,MDEyOklzc3VlQ29tbWVudDI4NDI3ODI0NA==,1217238,2017-03-06T00:49:22Z,2017-03-06T00:49:22Z,MEMBER,"> I didn't notice Dataset keeps the order of data_vars.

It's not a strict requirement, but we try to do it when possible because it leads to more predictable results.","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,211323643
https://github.com/pydata/xarray/pull/1289#issuecomment-284226244,https://api.github.com/repos/pydata/xarray/issues/1289,284226244,MDEyOklzc3VlQ29tbWVudDI4NDIyNjI0NA==,6815844,2017-03-05T13:00:02Z,2017-03-05T13:02:00Z,MEMBER,"+ Now DatasetRolling.reduce() keeps its original `data_vars` order. 
+ Dataset.rolling is added to api.rst.

I updated `DatasetRolling` so that its `reduce` method relies mostly on `DataArrayRolling.reduce` rather than `window.reduce`.
This change makes not only the reduced-Dataset keep its original order of `data_var`, but also makes the code much cleaner.
I moved many methods from `Rolling` to `DataArrayRolling` because `DatasetRolling` does not use them anymore.

However, I guess `Rolling` class is originally designed to be used also for `Dataset`, as discussed in #859. 
If someone has any requests, I will reconsider the class structure.","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,211323643
https://github.com/pydata/xarray/pull/1289#issuecomment-284217037,https://api.github.com/repos/pydata/xarray/issues/1289,284217037,MDEyOklzc3VlQ29tbWVudDI4NDIxNzAzNw==,6815844,2017-03-05T09:48:55Z,2017-03-05T09:48:55Z,MEMBER,"@shoyer Thanks for the review.

I didn't notice Dataset keeps the order of `data_vars`.
I think I should change the direction, because the data_var's order is destroyed after the final `merge` call.

Maybe `DatasetRolling` should not be a child of `Rolling`, but an independent class that has an `OrderedDict` of `Rolling` for each `data_var` to preserve the variable order. ","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,211323643
https://github.com/pydata/xarray/pull/1289#issuecomment-283931660,https://api.github.com/repos/pydata/xarray/issues/1289,283931660,MDEyOklzc3VlQ29tbWVudDI4MzkzMTY2MA==,6815844,2017-03-03T11:30:45Z,2017-03-03T11:30:45Z,MEMBER,"@MaximilianR 

Thanks for the review.
I made `rolling_cls` and `groupby_cls` private, and added some docstrings.

Also, I removed `_axis_num` property in Rolling class to make it compatible with Dataset.
It can be replaced by da.get_axis_num(self.dim).","{""total_count"": 1, ""+1"": 1, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,211323643