def test_distance_mdl_computation_mp(self):
mdl_slcs = eda.MDLSingleLabelClassifiedSamples(
self.x50x5, labs=self.labs50, encode_type="data",
mdl_method=eda.mdl.GKdeMdl, metric="euclidean")
data_no_lab_mdl = mdl_slcs.no_lab_mdl(nprocs=2)
mdl_slcs = eda.MDLSingleLabelClassifiedSamples(
self.x50x5, labs=self.labs50, encode_type="distance",
mdl_method=eda.mdl.GKdeMdl, metric="euclidean")
no_lab_mdl = mdl_slcs.no_lab_mdl(nprocs=2)
(ulab_mdl_sum, ulab_s_ind_l, ulab_cnt_l,
ulab_mdl_l, cluster_mdl) = mdl_slcs.lab_mdl(nprocs=2)
assert ulab_s_ind_l == [list(range(10)), list(range(10, 45)),
list(range(45, 50))]
assert ulab_mdl_sum == np.sum(ulab_mdl_l)
ulab_cnt_l = [10, 35, 5]
for i in range(3):
ci_mdl = eda.MDLSingleLabelClassifiedSamples(
mdl_slcs._x[ulab_s_ind_l[i]].copy(),
labs=[self.labs50[ii] for ii in ulab_s_ind_l[i]],
metric="euclidean", encode_type="distance",
mdl_method=eda.mdl.GKdeMdl)
ci_mdl_no_lab_res = ci_mdl.no_lab_mdl(nprocs=2)
np.testing.assert_allclose(
ci_mdl_no_lab_res,
ulab_mdl_l[i] - cluster_mdl * ulab_cnt_l[i] / 50)
def test_mdl_method(self):
zigk_mdl_slcs = eda.MDLSingleLabelClassifiedSamples(
self.x50x5, labs=self.labs50, mdl_method=eda.mdl.ZeroIGKdeMdl,
metric="euclidean")
gk_mdl_slcs = eda.MDLSingleLabelClassifiedSamples(
self.x50x5, labs=self.labs50, mdl_method=eda.mdl.GKdeMdl,
metric="euclidean")
assert gk_mdl_slcs._mdl_method is eda.mdl.GKdeMdl
assert zigk_mdl_slcs._mdl_method is eda.mdl.ZeroIGKdeMdl
assert zigk_mdl_slcs.no_lab_mdl() != gk_mdl_slcs.no_lab_mdl()
for mdl_method in eda.MDL_METHODS:
mdl_slcs = eda.MDLSingleLabelClassifiedSamples(
self.x50x5, labs=self.labs50, mdl_method=mdl_method,
metric="euclidean")
(ulab_mdl_sum, ulab_s_ind_l, ulab_cnt_l,
ulab_mdl_l, cluster_mdl) = mdl_slcs.lab_mdl()
assert ulab_s_ind_l == [list(range(10)), list(range(10, 45)),
list(range(45, 50))]
assert ulab_mdl_sum == np.sum(ulab_mdl_l)
ulab_cnt_l = [10, 35, 5]
for i in range(3):
ci_mdl = eda.MDLSingleLabelClassifiedSamples(
self.x50x5[ulab_s_ind_l[i], :],
labs=[self.labs50[ii] for ii in ulab_s_ind_l[i]],
mdl_method=mdl_method,
metric="euclidean")
np.testing.assert_allclose(
ci_mdl.no_lab_mdl(),
ulab_mdl_l[i] - cluster_mdl * ulab_cnt_l[i] / 50)
def test_auto_param(self):
eda.MDLSingleLabelClassifiedSamples(
self.x50x5, labs=self.labs50, encode_type="auto",
mdl_method=None, metric="euclidean")
eda.MDLSingleLabelClassifiedSamples(
np.zeros((100, 101)), labs=[0]*100, encode_type="auto",
mdl_method=None, metric="euclidean")
eda.MDLSingleLabelClassifiedSamples(
np.ones((100, 100)), labs=[0]*100, encode_type="auto",
mdl_method=None, metric="euclidean")
eda.MDLSingleLabelClassifiedSamples(
[[], []], labs=[0]*2, encode_type="auto",
mdl_method=None, metric="euclidean")
def test_wrong_encode_type(self):
with pytest.raises(ValueError) as excinfo:
eda.MDLSingleLabelClassifiedSamples(
self.x50x5, labs=self.labs50, encode_type="123",
metric="euclidean").no_lab_mdl()
with pytest.raises(ValueError) as excinfo:
eda.MDLSingleLabelClassifiedSamples(
self.x50x5, labs=self.labs50, encode_type=1,
metric="euclidean").no_lab_mdl()
with pytest.raises(ValueError) as excinfo:
eda.MDLSingleLabelClassifiedSamples(
self.x50x5, labs=self.labs50, encode_type=None,
metric="euclidean").no_lab_mdl()
def test_encode_mdl(self):
mdl_slcs = eda.MDLSingleLabelClassifiedSamples(
self.x50x5, labs=self.labs50, metric="euclidean")
# wrong dimensions
with pytest.raises(ValueError) as excinfo:
mdl_slcs.encode(np.zeros((10, 3)))
with pytest.raises(ValueError) as excinfo:
mdl_slcs.encode(np.zeros(20))
with pytest.raises(ValueError) as excinfo:
mdl_slcs.encode(np.zeros(20), col_summary_func=1)
with pytest.raises(ValueError) as excinfo:
mdl_slcs.encode(np.zeros(20), col_summary_func=None)
emdl = mdl_slcs.encode(np.arange(100).reshape(-1, 5))
emdl2 = mdl_slcs.encode(np.arange(100).reshape(-1, 5), nprocs=2)
np.testing.assert_approx_equal(emdl, emdl2)
emdl3 = eda.MDLSingleLabelClassifiedSamples(
self.x50x5, mdl_method=eda.mdl.GKdeMdl, labs=self.labs50,
metric="euclidean").encode(np.arange(100).reshape(-1, 5))
assert emdl != emdl3
emdl4 = eda.MDLSingleLabelClassifiedSamples(
np.zeros((50, 5)), mdl_method=eda.mdl.GKdeMdl, labs=self.labs50,
metric="euclidean").encode(np.arange(100).reshape(-1, 5))
assert emdl != emdl4
emdl5 = eda.MDLSingleLabelClassifiedSamples(
np.zeros((50, 5)), mdl_method=eda.mdl.GKdeMdl, labs=self.labs50,
metric="euclidean").encode(np.arange(100).reshape(-1, 5),
non_zero_only=True)
assert emdl5 != emdl3
emdl6 = eda.MDLSingleLabelClassifiedSamples(
self.x50x5, encode_type="distance", mdl_method=eda.mdl.GKdeMdl,
labs=self.labs50, metric="euclidean").encode(
np.arange(100).reshape(-1, 50), non_zero_only=True)
assert emdl5 != emdl3
def test_mdl_computation(self):
mdl_slcs = eda.MDLSingleLabelClassifiedSamples(
self.x50x5, labs=self.labs50, metric="euclidean")
no_lab_mdl = mdl_slcs.no_lab_mdl()
(ulab_mdl_sum, ulab_s_ind_l, ulab_cnt_l, ulab_mdl_l,
cluster_mdl) = mdl_slcs.lab_mdl()
assert ulab_s_ind_l == [list(range(10)), list(range(10, 45)),
list(range(45, 50))]
assert ulab_mdl_sum == np.sum(ulab_mdl_l)
ulab_cnt_l = [10, 35, 5]
for i in range(3):
ci_mdl = eda.MDLSingleLabelClassifiedSamples(
self.x50x5[ulab_s_ind_l[i], :],
labs=[self.labs50[ii] for ii in ulab_s_ind_l[i]],
metric="euclidean")
np.testing.assert_allclose(
ci_mdl.no_lab_mdl(),
ulab_mdl_l[i] - cluster_mdl * ulab_cnt_l[i] / 50)
def test_wrong_mdl_method(self):
with pytest.raises(ValueError) as excinfo:
eda.MDLSingleLabelClassifiedSamples(
self.x50x5, labs=self.labs50, mdl_method="123",
metric="euclidean").no_lab_mdl()
with pytest.raises(ValueError) as excinfo:
eda.MDLSingleLabelClassifiedSamples(
self.x50x5, labs=self.labs50, mdl_method=int,
metric="euclidean").no_lab_mdl()
with pytest.raises(ValueError) as excinfo:
eda.MDLSingleLabelClassifiedSamples(
self.x50x5, labs=self.labs50, mdl_method="ZeroIMdl",
metric="euclidean").no_lab_mdl()
with pytest.raises(ValueError) as excinfo:
eda.MDLSingleLabelClassifiedSamples(
self.x50x5, labs=self.labs50, mdl_method=2,
metric="euclidean").no_lab_mdl()
def test_lab_mdl_ret_internal(self):
mdl_slcs = eda.MDLSingleLabelClassifiedSamples(
self.x50x5, labs=self.labs50, metric="euclidean")
((ulab_mdl_sum, ulab_s_ind_l, ulab_cnt_l, ulab_mdl_l,
cluster_mdl), mdl_l) = mdl_slcs.lab_mdl(ret_internal=True)
np.testing.assert_allclose(sum(mdl_l) + cluster_mdl,
sum(ulab_mdl_l))
lab_mdl_res = mdl_slcs.lab_mdl()
ulab_mdl_sum2 = lab_mdl_res.ulab_mdl_sum
assert ulab_mdl_sum2 == ulab_mdl_sum
assert ulab_mdl_sum2 == np.sum(lab_mdl_res.ulab_mdls)
