def test_multidim_coord():
coord_1 = np.tile(['a'] * 51 + ['b'] * 49, (10, 1)).T
coord_2 = np.random.random((100, 10, 10))
X_ds = xr.Dataset(
{
'var_1':
(['sample', 'feat_1', 'feat_2'], np.random.random((100, 10, 10)))
},
coords={
'sample': range(100),
'feature': range(10),
'coord_1': (['sample', 'feat_1'], coord_1),
'coord_2': (['sample', 'feat_1', 'feat_2'], coord_2)
})
target_1 = Target(coord='coord_1',
transform_func=LabelBinarizer().fit_transform,
dim='sample')(X_ds)
target_2 = Target(coord='coord_2',
dim=['sample', 'feat_1'],
reduce_func=np.mean)(X_ds)
npt.assert_equal(target_1, LabelBinarizer().fit_transform(coord_1[:, 0]))
npt.assert_equal(target_2, np.mean(coord_2, 2))
def test_getitem():
coord_1 = ['a'] * 51 + ['b'] * 49
coord_2 = list(range(10)) * 10
X_ds = xr.Dataset(
{'var_1': (['sample', 'feature'], np.random.random((100, 10)))},
coords={
'sample': range(100),
'feature': range(10),
'coord_1': (['sample'], coord_1),
'coord_2': (['sample'], coord_2)
})
target = Target(coord='coord_1',
transform_func=LabelBinarizer().fit_transform)(X_ds)
y_test = target[-1]
assert y_test == LabelBinarizer().fit_transform(coord_1)[-1]
# test lazy eval
target = Target(coord='coord_1',
transform_func=LabelBinarizer().fit_transform,
lazy=True)(X_ds)
y_test = target[-1]
assert y_test == LabelBinarizer().fit_transform(coord_1)[-1]
assert not y_test.lazy
def test_getitem():
coord_1 = ["a"] * 51 + ["b"] * 49
coord_2 = list(range(10)) * 10
X_ds = xr.Dataset(
{"var_1": (["sample", "feature"], np.random.random((100, 10)))},
coords={
"sample": range(100),
"feature": range(10),
"coord_1": (["sample"], coord_1),
"coord_2": (["sample"], coord_2),
},
)
target = Target(coord="coord_1",
transform_func=LabelBinarizer().fit_transform)(X_ds)
y_test = target[-1]
assert y_test == LabelBinarizer().fit_transform(coord_1)[-1]
# test lazy eval
target = Target(
coord="coord_1",
transform_func=LabelBinarizer().fit_transform,
lazy=True,
)(X_ds)
y_test = target[-1]
assert y_test == LabelBinarizer().fit_transform(coord_1)[-1]
assert not y_test.lazy
def test_multidim_coord():
coord_1 = np.tile(["a"] * 51 + ["b"] * 49, (10, 1)).T
coord_2 = np.random.random((100, 10, 10))
X_ds = xr.Dataset(
{
"var_1": (
["sample", "feat_1", "feat_2"],
np.random.random((100, 10, 10)),
)
},
coords={
"sample": range(100),
"feature": range(10),
"coord_1": (["sample", "feat_1"], coord_1),
"coord_2": (["sample", "feat_1", "feat_2"], coord_2),
},
)
target_1 = Target(
coord="coord_1",
transform_func=LabelBinarizer().fit_transform,
dim="sample",
)(X_ds)
target_2 = Target(coord="coord_2",
dim=["sample", "feat_1"],
reduce_func=np.mean)(X_ds)
npt.assert_equal(target_1, LabelBinarizer().fit_transform(coord_1[:, 0]))
npt.assert_equal(target_2, np.mean(coord_2, 2))
def test_str():
assert str(Target()).startswith(
'Unassigned sklearn_xarray.Target without coordinate')
assert str(Target(coord='test')).startswith(
'Unassigned sklearn_xarray.Target with coordinate "test"')
assert str(Target()(
np.ones(10))).startswith('sklearn_xarray.Target with data:')
def test_is_target():
target = Target()
assert is_target(target)
not_a_target = 1
assert not is_target(not_a_target)
def test_constructor():
from sklearn_xarray.utils import convert_to_ndarray
coord_1 = ["a"] * 51 + ["b"] * 49
coord_2 = list(range(10)) * 10
X_ds = xr.Dataset(
{"var_1": (["sample", "feature"], np.random.random((100, 10)))},
coords={
"sample": range(100),
"feature": range(10),
"coord_1": (["sample"], coord_1),
"coord_2": (["sample"], coord_2),
},
)
target = Target(transform_func=convert_to_ndarray)
target.assign_to(X_ds)
npt.assert_equal(target.values, np.array(X_ds.var_1))
target = Target(coord="coord_1", transformer=LabelBinarizer())(X_ds)
npt.assert_equal(target.values, LabelBinarizer().fit_transform(coord_1))
def test_array():
coord_1 = ['a'] * 51 + ['b'] * 49
coord_2 = list(range(10)) * 10
X_ds = xr.Dataset(
{'var_1': (['sample', 'feature'], np.random.random((100, 10)))},
coords={
'sample': range(100),
'feature': range(10),
'coord_1': (['sample'], coord_1),
'coord_2': (['sample'], coord_2)
})
target = Target(coord='coord_1',
transform_func=LabelBinarizer().fit_transform,
lazy=True)(X_ds)
npt.assert_equal(np.array(target), LabelBinarizer().fit_transform(coord_1))
def test_shape_and_ndim():
coord_1 = ['a'] * 51 + ['b'] * 49
coord_2 = list(range(10)) * 10
X_ds = xr.Dataset(
{'var_1': (['sample', 'feature'], np.random.random((100, 10)))},
coords={
'sample': range(100),
'feature': range(10),
'coord_1': (['sample'], coord_1),
'coord_2': (['sample'], coord_2)
})
target = Target(coord='coord_1',
transform_func=LabelBinarizer().fit_transform)(X_ds)
npt.assert_equal(target.shape,
LabelBinarizer().fit_transform(coord_1).shape)
npt.assert_equal(target.ndim, LabelBinarizer().fit_transform(coord_1).ndim)
def test_array():
coord_1 = ["a"] * 51 + ["b"] * 49
coord_2 = list(range(10)) * 10
X_ds = xr.Dataset(
{"var_1": (["sample", "feature"], np.random.random((100, 10)))},
coords={
"sample": range(100),
"feature": range(10),
"coord_1": (["sample"], coord_1),
"coord_2": (["sample"], coord_2),
},
)
target = Target(
coord="coord_1",
transform_func=LabelBinarizer().fit_transform,
lazy=True,
)(X_ds)
npt.assert_equal(np.array(target), LabelBinarizer().fit_transform(coord_1))
def test_shape_and_ndim():
coord_1 = ["a"] * 51 + ["b"] * 49
coord_2 = list(range(10)) * 10
X_ds = xr.Dataset(
{"var_1": (["sample", "feature"], np.random.random((100, 10)))},
coords={
"sample": range(100),
"feature": range(10),
"coord_1": (["sample"], coord_1),
"coord_2": (["sample"], coord_2),
},
)
target = Target(coord="coord_1",
transform_func=LabelBinarizer().fit_transform)(X_ds)
npt.assert_equal(target.shape,
LabelBinarizer().fit_transform(coord_1).shape)
npt.assert_equal(target.ndim, LabelBinarizer().fit_transform(coord_1).ndim)
def test_constructor():
from sklearn_xarray.utils import convert_to_ndarray
coord_1 = ['a'] * 51 + ['b'] * 49
coord_2 = list(range(10)) * 10
X_ds = xr.Dataset(
{'var_1': (['sample', 'feature'], np.random.random((100, 10)))},
coords={
'sample': range(100),
'feature': range(10),
'coord_1': (['sample'], coord_1),
'coord_2': (['sample'], coord_2)
})
target = Target(transform_func=convert_to_ndarray)
target.assign_to(X_ds)
npt.assert_equal(target.values, np.array(X_ds.var_1))
target = Target(coord='coord_1', transformer=LabelBinarizer())(X_ds)
npt.assert_equal(target.values, LabelBinarizer().fit_transform(coord_1))
#
# .. tip::
#
# To use multi-processing, set ``n_jobs=-1``.
gs = GridSearchCV(
pl, cv=cv, n_jobs=1, verbose=1, param_grid={
'pca__n_components': [10, 20]
})
##############################################################################
# The label to classify is the activity which we convert to an integer
# representation for the classification.
y = Target(coord='activity',
transform_func=LabelEncoder().fit_transform,
dim='sample')(X)
##############################################################################
# Finally, we run the grid search and print out the best parameter combination.
if __name__ == '__main__': # in order for n_jobs=-1 to work on Windows
gs.fit(X, y)
print('Best parameters: {0}'.format(gs.best_params_))
print('Accuracy: {0}'.format(gs.best_score_))
##############################################################################
# .. note::
#
# The performance of this classifier is obviously pretty bad,
# it was chosen for execution speed, not accuracy.
def fit(self,
data,
labels,
clf,
grid_search=False,
targ_name='targ',
targ_dim_name='sample',
col=None):
"""
Fits a classifier given class labels
Args:
data (DataArray): The data to predict on.
labels (str | Path | GeoDataFrame): Class labels as polygon geometry.
clf (object): The classifier or classification pipeline.
grid_search (Optional[bool]): Whether to use cross-validation.
targ_name (Optional[str]): The target name.
targ_dim_name (Optional[str]): The target coordinate name.
col (Optional[str]): The column in ``labels`` you want to assign values from.
If ``None``, creates a binary raster.
Returns:
``xarray.DataArray``, ``object``:
Reshaped `data`, classifier object
Example:
>>> import geowombat as gw
>>> from geowombat.data import l8_224078_20200518, l8_224078_20200518_polygons
>>> from geowombat.ml import fit
>>>
>>> import geopandas as gpd
>>> from sklearn_xarray.preprocessing import Featurizer
>>> from sklearn.pipeline import Pipeline
>>> from sklearn.preprocessing import StandardScaler, LabelEncoder
>>> from sklearn.decomposition import PCA
>>> from sklearn.naive_bayes import GaussianNB
>>>
>>> le = LabelEncoder()
>>>
>>> labels = gpd.read_file(l8_224078_20200518_polygons)
>>> labels['lc'] = le.fit(labels.name).transform(labels.name)
>>>
>>> # Use a data pipeline
>>> pl = Pipeline([('featurizer', Featurizer()),
>>> ('scaler', StandardScaler()),
>>> ('pca', PCA()),
>>> ('clf', GaussianNB())])
>>>
>>> with gw.open(l8_224078_20200518) as src:
>>> X, clf = fit(src, labels, pl, grid_search=True, col='lc')
"""
data = self._prepare_labels(data, labels, col, targ_name)
X, Xna = self._prepare_predictors(data, targ_name)
clf = self._prepare_classifiers(clf)
if grid_search:
clf = self.grid_search_cv(clf)
# TODO: should we be using lazy=True?
y = Target(coord=targ_name,
transform_func=LabelEncoder().fit_transform,
dim=targ_dim_name)(Xna)
clf.fit(Xna, y)
return X, clf
# .. tip::
#
# To use multi-processing, set ``n_jobs=-1``.
gs = GridSearchCV(pl,
cv=cv,
n_jobs=1,
verbose=1,
param_grid={"pca__n_components": [10, 20]})
##############################################################################
# The label to classify is the activity which we convert to an integer
# representation for the classification.
y = Target(coord="activity",
transform_func=LabelEncoder().fit_transform,
dim="sample")(X)
##############################################################################
# Finally, we run the grid search and print out the best parameter combination.
if __name__ == "__main__": # in order for n_jobs=-1 to work on Windows
gs.fit(X, y)
print("Best parameters: {0}".format(gs.best_params_))
print("Accuracy: {0}".format(gs.best_score_))
##############################################################################
# .. note::
#
# The performance of this classifier is obviously pretty bad,
# it was chosen for execution speed, not accuracy.
