def test_kmeans_modelweights(self):
ds = xr.open_dataset(self.nc_file)
# Test sample weights
model_weights = np.ones(ds.data.shape[0])
model_weights[[4, 7, 10, 23]] = 0
# set to zero for some models that are selected in n_cluster test - these models should not be selected now
[ids, cluster, fig_data] = ensembles.kmeans_reduce_ensemble(
data=ds.data,
method={"n_clusters": 4},
random_state=42,
make_graph=False,
model_weights=model_weights,
)
for i in np.where(model_weights == 0)[0]:
# as long as the cluster has more than one member the models w/ weight==0 should not be present
if np.sum(cluster == cluster[i]) > 1:
assert i not in ids
model_weights = np.ones(ds.data.shape[0])
model_weights[[0, 3, 4, 6, 7, 10, 11, 12, 13]] = 0
[ids, cluster, fig_data] = ensembles.kmeans_reduce_ensemble(
data=ds.data,
method={"n_clusters": 9},
random_state=42,
make_graph=False,
model_weights=model_weights,
)
for i in np.where(model_weights == 0)[0]:
# as long as the cluster has more than one member the models w/ weight==0 should not be present
if np.sum(cluster == cluster[i]) > 1:
assert i not in ids
def test_kmeans_variweights(self):
pytest.importorskip("sklearn", minversion="0.24.1")
ds = open_dataset(self.nc_file)
# Test sample weights
var_weights = np.ones(ds.data.shape[1])
# reduce weights for some variables
var_weights[3:] = 0.25
[ids, cluster, fig_data] = ensembles.kmeans_reduce_ensemble(
data=ds.data,
method={"rsq_cutoff": 0.9},
random_state=42,
make_graph=False,
variable_weights=var_weights,
)
assert ids == [1, 3, 8, 10, 13, 14, 16, 19, 20]
assert len(ids) == 9
# using RSQ optimize and try zero weights
var_weights = np.ones(ds.data.shape[1])
var_weights[[1, 4]] = 0
[ids, cluster, fig_data] = ensembles.kmeans_reduce_ensemble(
data=ds.data,
method={"rsq_optimize": None},
random_state=42,
make_graph=False,
variable_weights=var_weights,
)
# Results here may change according to sklearn version, hence the *isin* intead of ==
assert all(np.isin([12, 13, 16], ids))
assert len(ids) == 6
def test_kmeans_sampleweights(self):
ds = open_dataset(self.nc_file)
# Test sample weights
sample_weights = np.ones(ds.data.shape[0])
# boost weights for some sims
sample_weights[[0, 20]] = 15
[ids, cluster, fig_data] = ensembles.kmeans_reduce_ensemble(
data=ds.data,
method={"rsq_cutoff": 0.5},
random_state=42,
make_graph=False,
sample_weights=sample_weights,
)
assert ids == [0, 20, 23]
assert len(ids) == 3
# RSQ optimize
sample_weights = np.ones(ds.data.shape[0])
# try zero weights
sample_weights[[6, 18, 22]] = 0
[ids, cluster, fig_data] = ensembles.kmeans_reduce_ensemble(
data=ds.data,
method={"rsq_optimize": None},
random_state=0,
make_graph=False,
sample_weights=sample_weights,
)
assert ids == [4, 5, 7, 10, 11, 12, 13]
assert len(ids) == 7
def test_kmeans_nclust(self):
ds = xr.open_dataset(self.nc_file)
[ids, cluster, fig_data] = ensembles.kmeans_reduce_ensemble(
data=ds.data, method={"n_clusters": 4}, random_state=42, make_graph=False
)
assert ids == [4, 7, 10, 23]
assert len(ids) == 4
[ids, cluster, fig_data] = ensembles.kmeans_reduce_ensemble(
ds.data, method={"n_clusters": 9}, random_state=42, make_graph=False
)
assert ids == [0, 3, 4, 6, 7, 10, 11, 12, 13]
assert len(ids) == 9
def test_kmeans_variweights(self):
pytest.importorskip("sklearn", minversion="0.22")
ds = xr.open_dataset(self.nc_file)
# Test sample weights
var_weights = np.ones(ds.data.shape[1])
# reduce weights for some variables
var_weights[3:] = 0.25
[ids, cluster, fig_data] = ensembles.kmeans_reduce_ensemble(
data=ds.data,
method={"rsq_cutoff": 0.9},
random_state=42,
make_graph=False,
variable_weights=var_weights,
)
assert ids == [1, 3, 8, 10, 13, 14, 16, 19, 20]
assert len(ids) == 9
# using RSQ optimize
[ids, cluster, fig_data] = ensembles.kmeans_reduce_ensemble(
data=ds.data,
method={"rsq_optimize": None},
random_state=42,
make_graph=False,
variable_weights=var_weights,
)
assert ids == [2, 4, 8, 13, 14, 22]
assert len(ids) == 6
# try zero weights
var_weights = np.ones(ds.data.shape[1])
var_weights[[1, 4]] = 0
[ids, cluster, fig_data] = ensembles.kmeans_reduce_ensemble(
data=ds.data,
method={"rsq_optimize": None},
random_state=42,
make_graph=False,
variable_weights=var_weights,
)
assert ids == [4, 10, 12, 13, 16]
assert len(ids) == 5
def test_kmeans_rsqopt(self):
pytest.importorskip("sklearn", minversion="0.24.1")
ds = open_dataset(self.nc_file)
[ids, cluster, fig_data] = ensembles.kmeans_reduce_ensemble(
data=ds.data,
method={"rsq_optimize": None},
random_state=42,
make_graph=False,
)
assert ids == [3, 4, 5, 7, 10, 11, 12, 13]
assert len(ids) == 8
def test_kmeans_rsqcutoff_with_graphs(self):
pytest.importorskip("sklearn", minversion="0.22")
ds = xr.open_dataset(self.nc_file)
# use random state variable to ensure consistent clustering in tests:
[ids, cluster, fig_data] = ensembles.kmeans_reduce_ensemble(
data=ds.data, method={"rsq_cutoff": 0.9}, random_state=42, make_graph=True
)
assert ids == [0, 1, 3, 4, 6, 7, 8, 9, 10, 11, 12, 15, 20, 22]
assert len(ids) == 14
# Test max cluster option
[ids, cluster, fig_data] = ensembles.kmeans_reduce_ensemble(
data=ds.data,
method={"rsq_cutoff": 0.9},
random_state=42,
make_graph=True,
max_clusters=10,
)
assert ids == [0, 1, 3, 4, 6, 7, 10, 11, 18, 20]
assert len(ids) == 10
def test_kmeans_rsqcutoff_with_graphs(self):
pytest.importorskip("sklearn", minversion="0.24.1")
ds = open_dataset(self.nc_file)
# use random state variable to ensure consistent clustering in tests:
[ids, cluster, fig_data
] = ensembles.kmeans_reduce_ensemble(data=ds.data,
method={"rsq_cutoff": 0.5},
random_state=42,
make_graph=True)
assert ids == [4, 7, 10, 23]
assert len(ids) == 4
# Test max cluster option
[ids, cluster, fig_data] = ensembles.kmeans_reduce_ensemble(
data=ds.data,
method={"rsq_cutoff": 0.5},
random_state=42,
make_graph=True,
max_clusters=3,
)
assert ids == [4, 7, 23]
assert len(ids) == 3
