[Bug] r.learn.ml- onehot encoding not working any more with scikit-learn 0.22 about grass-addons

[veroandreo]

Name of the addon
r.learn.m

Describe the bug
In the new scikit-learn 0.22 the OneHotEncode() function parameters are changed and category maps can not be used in r.learn.m current version.

To Reproduce
Test with category maps.

Please provide a set of commands to reproduce (you can use the example in the manual as a basis)

Expected behavior

Error message here
clf = Pipeline([('onehot', OneHotEncoder(
categorical_features=categorymaps

Additional context
There is a new version of r.learn.ml (separate train and predict modules) in github where the problem is solved. In this version (training and predicting in one module) the message could be modified to indicate the last version of python3-scikit-learn that is compatible with the script:
gs.fatal("Package python3-scikit-learn 0.18 or newer is not installed") .

Would you please provide the link where this new version of r.learn.ml exists?

[maitlang1]

Sorry, forgot the link to Steven's repo:
https://github.com/stevenpawley/r.learn.ml .

[neteler]

Let's keep it open since the problem isn't solved yet

[neteler]

note: @stevenpawley told me off-list that he prepares an update of r.learn.ml

[neteler]

@maitlang1 could you give #126 a try (New version of r.learn.ml)?

[maitlang1]

GRASS 7.9.dev (25m_raster):~ > g.extension -s r.learn.ml url=github.com/stevenpawley/r.learn.ml
WARNING: Extension <r.learn.ml> already installed. Re-installing...
Fetching <r.learn.ml> from
https://github.com/stevenpawley/r.learn.ml/archive/master.zip (be
patient)...
Compiling...
Installing...
Updating private addons metadata file...
WARNING: No addons metadata available. Addons metadata file not updated.
Installation of <r.learn.ml> successfully finished

The test

Extracting training data
Traceback (most recent call last):
File "/usr/local/grass79/scripts/r.learn.train", line 876,
in
main()
File "/usr/local/grass79/scripts/r.learn.train", line 588,
in main
X, y, cat = stack.extract_points(training_points, field)
File "/usr/local/grass79/etc/r.learn.ml/raster.py", line
886, in extract_points
self._grass_sql_dtype_to_numpy(df_cols[field])
File "/usr/local/grass79/etc/r.learn.ml/raster.py", line
815, in _grass_sql_dtype_to_numpy
return grass_vector_dtypes[dtype]
KeyError: 'int'

The response column description:
column:PE
description:
type:INTEGER
len:20
scale:0
precision:0
default:
nullok:yes
select:?
update:?

[stevenpawley]

Thanks for this, it should be fixed now in 0aa72e9

[maitlang1]

I compiled a grass database for testing r.learn.ml addon: https://drive.google.com/file/d/1wQEwJNKtFheD5cBoXQwdiuBaqk1bI-25/view?usp=sharing . The database structure description is included.

[stevenpawley]

I'm going to close this because I've switched all further work to r.learn.ml2