fix: Ml nodata (#310)

* simplify mosaic procedure

* adding mosiac and save unit tests

* fix path tifs

* resolve ml install issues

* handle 0s in X[targ_name]

* add todo

* reverting to main

* pin scikit fix

* add tests

* throw error if stacked by time
resolves #311

* droping test - meaningless output

---------

Co-authored-by: jgrss <jbgraesser@gmail.com>

setup.cfg

@@ -83,9 +83,10 @@ coreg = earthpy
 zarr = zarr
     numcodecs
 ml = dask-ml>=2022.5.27
-    scikit-learn>=0.23.0,<=1.2.0
+    scikit-learn==1.2.0
     lightgbm
-    sklearn-xarray@git+https://github.com/jgrss/sklearn-xarray.git
+    sklearn-xarray@git+https://github.com/mmann1123/sklearn-xarray.git
+    numpy_groupies
 perf = rtree
     pygeos
     netCDF4

src/geowombat/backends/xarray_.py

@@ -476,6 +476,7 @@ def reduce_func(
                 xr.where(left != tmp_nodata, left, right),
             )
 
+
     # Open all the data pointers
     data_arrays = [
         open_rasterio(
@@ -547,9 +548,9 @@ def reduce_func(
         attrs.update(tags)
         darray = darray.assign_attrs(**attrs)
 
+
     if dtype is not None:
         attrs = darray.attrs.copy()
-
         return darray.astype(dtype).assign_attrs(**attrs)
 
     else:

src/geowombat/ml/classifiers.py

@@ -88,8 +88,11 @@ def _prepare_predictors(self, data, targ_name):
 
         # drop nans
         try:
+            # prep target axis 
             Xna = X[~X[targ_name].isnull()]
-            Xna = X[X[targ_name] != X.gw.nodataval]
+            Xna = X[X[targ_name] != 0]  # Xtarg is being generated with meaningless 0s
+            # TODO: if X.gw.nodataval is not None:
+            #     Xna = X[X!= X.gw.nodata ]  # changes here would have to be reflected in y as well
         except KeyError:
             Xna = X
 
@@ -250,6 +253,11 @@ def fit(
             >>> with gw.open(l8_224078_20200518) as src:
             >>>    X, Xy, clf = fit(src, cl)
         """
+        if data.gw.has_time_coord:
+            # throw error
+            raise ValueError(
+                "DataArray must not have a time coordinate. Use stack_dim='band' with gw.open() or use .isel(time=0) to select a single time slice."
+            )
         if clf._estimator_type == "clusterer":
             data = self._add_time_dim(data)
             X, Xna = self._prepare_predictors(data, targ_name)

tests/ml_test.py

@@ -112,12 +112,39 @@ def test_tree_predict(self):
         self.assertTrue(np.all(np.isnan(y2.values[0, 0:5, 0])))
         self.assertTrue(
             np.allclose(
-                y1.values[0, -5:-1, 0],
-                y2.values[0, -5:-1, 0],
+                y1.values,
+                y2.values,
                 equal_nan=True,
             )
         )
 
+    def test_tree_predict_nodata(self):
+        with gw.config.update(
+            ref_res=300,
+        ):
+            # assigning invalid nodata value
+            with gw.open(l8_224078_20200518, nodata=10) as src:
+                with warnings.catch_warnings():
+                    warnings.simplefilter(
+                        "ignore",
+                        (DeprecationWarning, FutureWarning, UserWarning),
+                    )
+                    y1 = fit_predict(src, tree_pipeline, aoi_poly, col="lc")
+            with gw.open(l8_224078_20200518) as src:
+                with warnings.catch_warnings():
+                    warnings.simplefilter(
+                        "ignore",
+                        (DeprecationWarning, FutureWarning, UserWarning),
+                    )
+                    y2 = fit_predict(src, tree_pipeline, aoi_poly, col="lc")
+
+        self.assertTrue(
+            np.allclose(
+                y1.values,
+                y2.values,
+                equal_nan=True,
+            )
+        )
     def test_output_type_attri(self):
 
         with gw.config.update(
@@ -135,7 +162,6 @@ def test_output_type_attri(self):
 
         self.assertTrue(isinstance(y1, xr_da))
         self.assertTrue(isinstance(y2, xr_da))
-        # self.assertTrue(isinstance(y1.chunks, tuple))
         self.assertTrue(len(y1.attrs) > 0)
         self.assertTrue(len(y2.attrs) > 0)
 
@@ -156,29 +182,6 @@ def test_fitpredict_eq_fit_predict_point(self):
 
         self.assertTrue(np.allclose(y1.values, y2.values, equal_nan=True))
 
-    def test_fitpredict_time_point(self):
-
-        with gw.config.update(
-            ref_res=300,
-        ):
-            with gw.open(
-                [l8_224078_20200518, l8_224078_20200518], stack_dim="time"
-            ) as src:
-                with warnings.catch_warnings():
-                    warnings.simplefilter(
-                        "ignore",
-                        (DeprecationWarning, FutureWarning, UserWarning),
-                    )
-                    y1 = fit_predict(
-                        src,
-                        pl_wo_feat,
-                        aoi_point,
-                        col="lc",
-                        mask_nodataval=False,
-                    )
-
-        self.assertTrue(np.all(y1.sel(time=1).values == y1.sel(time=2).values))
-
     def test_fitpredict_eq_fit_predict_cluster(self):
 
         with gw.config.update(
@@ -249,6 +252,29 @@ def test_classes_match_prediction_b(self):
             )
         )
 
+    # def test_fitpredict_time_point(self):
+
+    #     with gw.config.update(
+    #         ref_res=300,
+    #     ):
+    #         with gw.open(
+    #             [l8_224078_20200518, l8_224078_20200518], stack_dim="time"
+    #         ) as src:
+    #             with warnings.catch_warnings():
+    #                 warnings.simplefilter(
+    #                     "ignore",
+    #                     (DeprecationWarning, FutureWarning, UserWarning),
+    #                 )
+    #                 y1 = fit_predict(
+    #                     src,
+    #                     pl_wo_feat,
+    #                     aoi_point,
+    #                     col="lc",
+    #                     mask_nodataval=False,
+    #                 )
+
+    #     self.assertTrue(np.all(y1.sel(time=1).values == y1.sel(time=2).values))
+
     # def test_nodataval_replace(self):
 
     #     with gw.config.update(ref_res=300):