fix: Origin/mmann1123 ml nodata

[mmann1123]

What is this PR changing?

• Dropping missing before fitting model
• corrected output dtype - now DataArray with attributes of source

• added tests of output dtype, attributes, and classification output values

  closes #193

[mmann1123]

@jgrss CI hates me. The tests are consistently working fine on my computer. Can you pull and test locally? Not sure what to do other than drop the tests.

[jgrss]

@jgrss CI hates me. The tests are consistently working fine on my computer. Can you pull and test locally? Not sure what to do other than drop the tests.

This is my fault. There is a cache setting in the CI. GeoWombat only gets re-built during tests if setup.cfg changes. I can remove this if it's annoying. For now, if you make any code changes that require a new install of geowombat, any minor change in setup.cfg will trigger it.

[jgrss]

@mmann1123 Apologies, out of habit I merged this after approving! My comments were minor but see if they warrant another PR, particularly the

Xna = X.gw.mask_nodata()

suggestion.

[mmann1123]

Yeah that might be helpful.

On Wed, Sep 21, 2022 at 6:49 PM Jordan Graesser @.***> wrote:

@.**** commented on this pull request.

In src/geowombat/ml/classifiers.py https://github.com/jgrss/geowombat/pull/207#discussion_r977045863:

 def _prepare_predictors(self, data, targ_name):

     X = self._stack_it(data)

     # drop nans
     try:
         Xna = X[~X[targ_name].isnull()]

• Xna = X[X[targ_name] != X.attrs["nodatavals"][0]]

How about we add a property called nodata?

data.gw.nodata

— Reply to this email directly, view it on GitHub https://github.com/jgrss/geowombat/pull/207#discussion_r977045863, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABHR6VDGWMCS5PD35QFVZWTV7OGHPANCNFSM6AAAAAAQRH5MKA . You are receiving this because you were mentioned.Message ID: @.***>