Netcdf improvements

[jessicaaustin]

• Improve documentation and tests for NcQcConfig
• Update NcQcConfig save_to_netcdf method to make it compliant with latest CF and IOOS metadata profiles
  • See Proposal: Add QARTOD quality flag names to standard name list cf-convention/cf-conventions#216
  • Will also fix NcQcConfig Object Ancillary Variable Issues #14
• Add option for NcQcConfig to generate aggregate flag (see QcConfig does not create an aggregate flag #15)
• Add netCDF notebook example (based on Netcdf examples #18)

note: i based this off of gen_agg_take2, so that would be merged into master before this PR. and if the approach for generating the aggregate flag changes, then this branch would need updates as well

[review-notebook-app]

Check out this pull request on 

You'll be able to see Jupyter notebook diff and discuss changes. Powered by ReviewNB.

[jessicaaustin]

note: i based this off of gen_agg_take2, so that would be merged into master before this PR. and if the approach for generating the aggregate flag changes, then this branch would need updates as well

[jessicaaustin]

@kwilcox could you please review?

In particular, these are some changes I'd love to get feedback on:

• I updated the NcQcConfig constructor to add arguments to indicate the name of the dimensions in the netcdf file
  • this is useful so you don't have to extract that data out yourself when calling run() -- it can just use the data from the proper variables in the netcdf file
  • see TestRunNcConfigClimatology for an example of the difference in approach here
  • there are a lot of ways to solve this same problem; curious if this approach feels pythonic/netcdf-y or not
• In save_to_netcdf, I added a fill value attribute to the qc variable. Seems reasonable?
• In save_to_netcdf, I removed valid_min and valid_max attributes. Does it make sense to have those for a list of flags anyway? (as opposed to a data variable which is continuous)
• To store the proper standard_name with the variable, I added an annotation, add_metadata, to each of the test functions

[kwilcox]

I updated the NcQcConfig constructor to add arguments to indicate the name of the dimensions in the netcdf file, this is useful so you don't have to extract that data out yourself when calling run() -- it can just use the data from the proper variables in the netcdf file.

I like it, wish this could be represented in the NcQcConfig file as well, but for now this is helpful. Would also be nice to default to the CF convention (extract the time and z axes if they are not provided).

In save_to_netcdf, I added a fill value attribute to the qc variable. Seems reasonable?

There was some discussion if a flag value of UNKNOWN should be filled/masked. Do you remember where the convo happened and what the outcome was? Would you assume that if the quality flag was UNKNOWN it should be masked? The opposing side would be to only mask the quality flag when the data was masked (we do that now). Right now: data input mask and the quality flag mask are always be equal. After PR: the quality flag mask would include input data mask and all UNKNOWN).

In save_to_netcdf, I removed valid_min and valid_max attributes. Does it make sense to have those for a list of flags anyway? (as opposed to a data variable which is continuous)

Yeah, this is still useful. Some tools use it for color bar bounds and other things. They could pull the info from flag_values, but this makes it more generic. I'll add back in unless you object.

To store the proper standard_name with the variable, I added an annotation, add_metadata, to each of the test functions

This is cool, works great for the purpose.

[jessicaaustin]

I updated the NcQcConfig constructor to add arguments to indicate the name of the dimensions in the netcdf file, this is useful so you don't have to extract that data out yourself when calling run() -- it can just use the data from the proper variables in the netcdf file.

I like it, wish this could be represented in the NcQcConfig file as well, but for now this is helpful. Would also be nice to default to the CF convention (extract the time and z axes if they are not provided).

Agreed. I'll add an issue for that to track it.

In save_to_netcdf, I added a fill value attribute to the qc variable. Seems reasonable?

There was some discussion if a flag value of UNKNOWN should be filled/masked. Do you remember where the convo happened and what the outcome was? Would you assume that if the quality flag was UNKNOWN it should be masked? The opposing side would be to only mask the quality flag when the data was masked (we do that now). Right now: data input mask and the quality flag mask are always be equal. After PR: the quality flag mask would include input data mask and all UNKNOWN).

Hmmmm I see what you mean. When you put it that way, I think "data input mask and the quality flag mask are always be equal" is the correct behavior. So I will revert my changes.

In save_to_netcdf, I removed valid_min and valid_max attributes. Does it make sense to have those for a list of flags anyway? (as opposed to a data variable which is continuous)

Yeah, this is still useful. Some tools use it for color bar bounds and other things. They could pull the info from flag_values, but this makes it more generic. I'll add back in unless you object.

Makes sense, thanks for the explanation.

[kwilcox]

I'm ok with whatever you decide on the fill_value, good to go after that