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/issues/3884#issuecomment-604838173,https://api.github.com/repos/pydata/xarray/issues/3884,604838173,MDEyOklzc3VlQ29tbWVudDYwNDgzODE3Mw==,1217238,2020-03-27T06:31:08Z,2020-03-27T06:31:08Z,MEMBER,"I wouldn’t worry too much about reusing bottleneck here, unless we really these functions will be the bottleneck in user code :)","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,587062505
https://github.com/pydata/xarray/issues/3884#issuecomment-604775165,https://api.github.com/repos/pydata/xarray/issues/3884,604775165,MDEyOklzc3VlQ29tbWVudDYwNDc3NTE2NQ==,2272878,2020-03-27T01:57:44Z,2020-03-27T01:57:44Z,CONTRIBUTOR,"@shoyer xarray uses bottleneck for that if it can in [`xarray.nputils`](https://github.com/pydata/xarray/blob/6378a711d50ba7f1ba9b2a451d4d1f5e1fb37353/xarray/core/nputils.py#L197), so copying the numpy method would result in a performance hit.  However, xarray maintains a wrapper around the numpy/bottleneck version in [`xarray.nanops`](https://github.com/pydata/xarray/blob/b3d3b4480b7fb63402eb6c02103bb8d6c7dbf93a/xarray/core/nanops.py#L101) where this could perhaps be implemented.","{""total_count"": 1, ""+1"": 1, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,587062505
https://github.com/pydata/xarray/issues/3884#issuecomment-604225711,https://api.github.com/repos/pydata/xarray/issues/3884,604225711,MDEyOklzc3VlQ29tbWVudDYwNDIyNTcxMQ==,1217238,2020-03-26T04:42:37Z,2020-03-26T04:42:37Z,MEMBER,"NumPy implements `nanargmax`, including raising the error, in Python. It would be very doable to copy a modified version into xarray.","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,587062505
https://github.com/pydata/xarray/issues/3884#issuecomment-604220730,https://api.github.com/repos/pydata/xarray/issues/3884,604220730,MDEyOklzc3VlQ29tbWVudDYwNDIyMDczMA==,2272878,2020-03-26T04:22:18Z,2020-03-26T04:22:18Z,CONTRIBUTOR,"The problem I had when implementing `idxmin` and `idxmax` is that this behavior is defined by `numpy`, not by xarray, and bottleneck follows the same behavior, with xarray generally delegating the computation to one of these.  So you would need to somehow work around the behavior of numpy in xarray or get a fix implemented both in numpy and bottleneck.","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,587062505
https://github.com/pydata/xarray/issues/3884#issuecomment-604186105,https://api.github.com/repos/pydata/xarray/issues/3884,604186105,MDEyOklzc3VlQ29tbWVudDYwNDE4NjEwNQ==,1217238,2020-03-26T02:07:27Z,2020-03-26T02:07:27Z,MEMBER,"The main concern here is _type stability_. Normally the return value of `argmax` is an integer dtype array, but NaN isn't a valid integer :(

My suggestion would be to add an optional `fill_value` argument, similar to the API under discussion for `idxmax` in https://github.com/pydata/xarray/pull/3871. If `fill_value` is specified (e.g., with `fill_value=np.nan` or `fill_value=-1`) then missing values are returned with the fill value instead of raising an error.","{""total_count"": 2, ""+1"": 2, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,587062505
https://github.com/pydata/xarray/issues/3884#issuecomment-603341320,https://api.github.com/repos/pydata/xarray/issues/3884,603341320,MDEyOklzc3VlQ29tbWVudDYwMzM0MTMyMA==,5635139,2020-03-24T16:18:54Z,2020-03-24T16:18:54Z,MEMBER,I think this would be a reasonable change,"{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,587062505