def run_test_impute_features(self, nprocs, use_hnsw):
tsdm = eda.SampleDistanceMatrix(spsp.csr_matrix(self.tx),
metric="euclidean",
use_pdist=False)
tsdm.s_knn_connectivity_matrix(5, use_hnsw=use_hnsw)
fmp = knn.FeatureImputation(tsdm)
res_sdml = fmp.impute_features([1, 3, 5], [1, 1, 3], [0.5, 1.5, 0.5],
[1, 1, 5],
nprocs=nprocs)
assert len(fmp._res_lut) == 3
assert res_sdml[0]._x.A.shape == (9, 2)
assert res_sdml[1]._x.A.shape == (9, 2)
assert res_sdml[2]._x.A.shape == (9, 2)
# value
ref_res0 = np.array([[1, 1], [1, 1], [1, 2], [2, 3], [2, 5], [0, 100],
[0, 100], [0, 101], [10, 100]])
np.testing.assert_equal(res_sdml[0]._x.A, ref_res0)
ref_res1 = np.array([[np.median([1, 1, 2]),
np.median([1, 2, 3])],
[np.median([1, 1, 2]),
np.median([1, 2, 3])], [np.median([1, 1, 2]), 2],
[2, 3], [2, 5], [np.median([0, 0, 10]), 100],
[np.median([0, 0, 10]), 100],
[np.median([0, 0, 10]), 101], [10, 100]])
np.testing.assert_equal(res_sdml[1]._x.A, ref_res1)
assert res_sdml[2]._x.A[0, 0] > 0
assert res_sdml[2]._x.A[0, 1] > 0
assert res_sdml[2]._x.A[5, 0] > 0
assert res_sdml[2]._x.A[6, 0] > 0
assert res_sdml[2]._x.A[7, 0] > 0
# lookup
np.testing.assert_equal(fmp._res_lut[(1, 1, 0.5, 1, np.median)][0].A,
res_sdml[0]._x.A)
np.testing.assert_equal(fmp._res_lut[(3, 1, 1.5, 1, np.median)][0].A,
res_sdml[1]._x.A)
np.testing.assert_equal(fmp._res_lut[(5, 3, 0.5, 5, np.median)][0].A,
res_sdml[2]._x.A)
fmp.impute_features([1, 3, 5], [1, 1, 3], [0.5, 1.5, 0.5], [1, 1, 5],
nprocs=nprocs)
assert len(fmp._res_lut) == 3
np.testing.assert_equal(fmp._sdm.x.A.tolist(), self.tx)
# run results should be placed with the correct order
fmp2 = knn.FeatureImputation(tsdm)
res1 = fmp2.impute_features([1, 3, 5], [1, 1, 3], [0.5, 1.5, 0.5],
[1, 1, 5],
nprocs=nprocs)
fmp3 = knn.FeatureImputation(tsdm)
res2 = fmp3.impute_features([2, 3, 4], [1, 1, 3], [0.5, 1.5, 0.5],
[1, 1, 5],
nprocs=nprocs)
res3 = fmp2.impute_features([2, 3, 4], [1, 1, 3], [0.5, 1.5, 0.5],
[1, 1, 5],
nprocs=nprocs)
np.testing.assert_equal(res1[1]._x.A, res3[1]._x.A)
np.testing.assert_equal(res2[0]._x.A, res3[0]._x.A)
np.testing.assert_equal(res2[2]._x.A, res3[2]._x.A)
def test_impute_features_single_run(self):
fmp = knn.FeatureImputation(self.tsdm)
d = fmp._sdm._d.copy()
assert d.shape == (9, 9)
res_sdml = fmp.impute_features([8], [3], [0.5], [10])
fmp2 = knn.FeatureImputation(self.tsdm)
res_sdml2 = fmp2.impute_features(8, 3, 0.5, 10)
np.testing.assert_equal(res_sdml2[0]._x, res_sdml[0]._x)
def test_impute_features_stat_fun(self):
fmp = knn.FeatureImputation(self.tsdm)
d = fmp._sdm._d.copy()
assert d.shape == (9, 9)
res_sdml = fmp.impute_features([8], [3], [0.5], [10])
res_sdml2 = fmp.impute_features(8, 3, 0.5, 10, 1, np.median)
np.testing.assert_equal(res_sdml2[0]._x, res_sdml[0]._x)
assert len(fmp._res_lut) == 1
res_sdml3 = fmp.impute_features(
8, 3, 0.5, 10, 1, lambda x, axis=0: np.median(x, axis=axis))
np.testing.assert_equal(res_sdml3[0]._x, res_sdml[0]._x)
assert len(fmp._res_lut) == 2
res_sdml4 = fmp.impute_features(
8, 3, 0.5, 10, 1, lambda x, axis=0: np.median(x, axis=axis))
np.testing.assert_equal(res_sdml4[0]._x, res_sdml[0]._x)
assert len(fmp._res_lut) == 3
res_sdml5 = fmp.impute_features(8, 3, 0.5, 10, 1, np.min)
assert not np.array_equal(res_sdml5[0]._x, res_sdml[0]._x)
assert len(fmp._res_lut) == 4
res_sdml6 = fmp.impute_features(8,
3,
0.5,
10,
1,
lambda x, axis=0: np.min(x, axis=axis))
np.testing.assert_equal(res_sdml5[0]._x, res_sdml6[0]._x)
assert len(fmp._res_lut) == 5
def test_impute_features_stat_fun(self):
tsdm = eda.SampleDistanceMatrix(spsp.csr_matrix(self.tx),
metric="euclidean",
use_pdist=False)
tsdm.s_knn_connectivity_matrix(5, use_hnsw=False)
fmp = knn.FeatureImputation(tsdm)
res_sdml = fmp.impute_features([8], [3], [0.5], [10])
res_sdml2 = fmp.impute_features(8, 3, 0.5, 10, 1, np.median)
np.testing.assert_equal(res_sdml2[0]._x.A, res_sdml[0]._x.A)
assert len(fmp._res_lut) == 1
res_sdml3 = fmp.impute_features(
8, 3, 0.5, 10, 1, lambda x, axis=0: np.median(x, axis=axis))
np.testing.assert_equal(res_sdml3[0]._x.A, res_sdml[0]._x.A)
assert len(fmp._res_lut) == 2
res_sdml4 = fmp.impute_features(
8, 3, 0.5, 10, 1, lambda x, axis=0: np.median(x, axis=axis))
np.testing.assert_equal(res_sdml4[0]._x.A, res_sdml[0]._x.A)
assert len(fmp._res_lut) == 3
res_sdml5 = fmp.impute_features(8, 3, 0.5, 10, 1, np.min)
assert not np.array_equal(res_sdml5[0]._x.A, res_sdml[0]._x.A)
assert len(fmp._res_lut) == 4
res_sdml6 = fmp.impute_features(8,
3,
0.5,
10,
1,
lambda x, axis=0: np.min(x, axis=axis))
np.testing.assert_equal(res_sdml5[0]._x.A, res_sdml6[0]._x.A)
assert len(fmp._res_lut) == 5
def test_impute_features_single_run(self):
tsdm = eda.SampleDistanceMatrix(spsp.csr_matrix(self.tx),
metric="euclidean",
use_pdist=False)
tsdm.s_knn_connectivity_matrix(5, use_hnsw=False)
fmp = knn.FeatureImputation(tsdm)
res_sdml = fmp.impute_features([8], [3], [0.5], [10])
fmp2 = knn.FeatureImputation(tsdm)
res_sdml2 = fmp2.impute_features(8, 3, 0.5, 10)
np.testing.assert_equal(res_sdml2[0]._x.A, res_sdml[0]._x.A)
tsdm = eda.SampleDistanceMatrix(spsp.csr_matrix(self.tx),
metric="euclidean",
use_pdist=False)
tsdm.s_knn_connectivity_matrix(5, use_hnsw=True)
fmp = knn.FeatureImputation(tsdm)
res_sdml = fmp.impute_features([8], [3], [0.5], [10])
fmp2 = knn.FeatureImputation(tsdm)
res_sdml2 = fmp2.impute_features(8, 3, 0.5, 10, verbose=True)
np.testing.assert_equal(res_sdml2[0]._x.A, res_sdml[0]._x.A)
def test_impute_features_wrong_args(self):
fmp = knn.FeatureImputation(self.tsdm)
# Invalid parameters
with pytest.raises(ValueError) as excinfo:
fmp.impute_features(0, 1, 10, 1)
with pytest.raises(ValueError) as excinfo:
fmp.impute_features(1, 0, 10, 1)
with pytest.raises(ValueError) as excinfo:
fmp.impute_features(1, 1, 10, 0)
with pytest.raises(ValueError) as excinfo:
fmp.impute_features(1, 1, 10, 0.5)
with pytest.raises(ValueError) as excinfo:
fmp.impute_features(0.5, 1, 10, 1)
with pytest.raises(ValueError) as excinfo:
fmp.impute_features(1, -0.1, 10, 1)
with pytest.raises(ValueError) as excinfo:
fmp.impute_features(1, 1, 1, 10, 0.5)
with pytest.raises(ValueError) as excinfo:
fmp.impute_features(9, 1, 1, 10, 1)
# invalid stats funcs
with pytest.raises(ValueError) as excinfo:
fmp.impute_features([1, 3, 5], [1, 1, 3],
[0.5, 1.5, 0.5], [1, 1, 5],
1, 1)
with pytest.raises(ValueError) as excinfo:
fmp.impute_features([1, 3, 5], [1, 1, 3],
[0.5, 1.5, 0.5], [1, 1, 5],
1, np.array)
with pytest.raises(ValueError) as excinfo:
fmp.impute_features([1, 3, 5], [1, 1, 3],
[0.5, 1.5, 0.5], [1, 1, 5],
1, lambda x, y: x + y)
with pytest.raises(ValueError) as excinfo:
fmp.impute_features([1, 3, 5], [1, 1, 3],
[0.5, 1.5, 0.5], [1, 1, 5],
1, lambda x, y: x + y)
# Parameters of different length
with pytest.raises(ValueError) as excinfo:
fmp.impute_features([1, 2], 1, 10, 1)
with pytest.raises(ValueError) as excinfo:
fmp.impute_features(1, [1, 2], 10, 1)
with pytest.raises(ValueError) as excinfo:
fmp.impute_features(1, 1, 10, [1, 2])
with pytest.raises(ValueError) as excinfo:
fmp.impute_features(1, 1, [10, 20], 1)
def test_impute_features_wrong_args(self):
tsdm = eda.SampleDistanceMatrix(spsp.csr_matrix(self.tx),
metric="euclidean",
use_pdist=False)
tsdm.s_knn_connectivity_matrix(5, use_hnsw=False)
fmp = knn.FeatureImputation(tsdm)
# Invalid parameters
with pytest.raises(ValueError) as excinfo:
fmp.impute_features(0, 1, 10, 1)
with pytest.raises(ValueError) as excinfo:
fmp.impute_features(1, 0, 10, 1)
with pytest.raises(ValueError) as excinfo:
fmp.impute_features(1, 1, 10, 0)
with pytest.raises(ValueError) as excinfo:
fmp.impute_features(1, 1, 10, 0.5)
with pytest.raises(ValueError) as excinfo:
fmp.impute_features(0.5, 1, 10, 1)
with pytest.raises(ValueError) as excinfo:
fmp.impute_features(1, -0.1, 10, 1)
with pytest.raises(ValueError) as excinfo:
fmp.impute_features(1, 1, 1, 10, 0.5)
with pytest.raises(ValueError) as excinfo:
fmp.impute_features(9, 1, 1, 10, 1)
# invalid stats funcs
with pytest.raises(ValueError) as excinfo:
fmp.impute_features([1, 3, 5], [1, 1, 3], [0.5, 1.5, 0.5],
[1, 1, 5], 1, 1)
with pytest.raises(ValueError) as excinfo:
fmp.impute_features([1, 3, 5], [1, 1, 3], [0.5, 1.5, 0.5],
[1, 1, 5], 1, np.array)
with pytest.raises(ValueError) as excinfo:
fmp.impute_features([1, 3, 5], [1, 1, 3], [0.5, 1.5, 0.5],
[1, 1, 5], 1, lambda x, y: x + y)
with pytest.raises(ValueError) as excinfo:
fmp.impute_features([1, 3, 5], [1, 1, 3], [0.5, 1.5, 0.5],
[1, 1, 5], 1, lambda x, y: x + y)
# Parameters of different length
with pytest.raises(ValueError) as excinfo:
fmp.impute_features([1, 2], 1, 10, 1)
with pytest.raises(ValueError) as excinfo:
fmp.impute_features(1, [1, 2], 10, 1)
with pytest.raises(ValueError) as excinfo:
fmp.impute_features(1, 1, 10, [1, 2])
with pytest.raises(ValueError) as excinfo:
fmp.impute_features(1, 1, [10, 20], 1)
def run_test_impute_features(self, nprocs):
fmp = knn.FeatureImputation(self.tsdm)
d = fmp._sdm._d.copy()
assert d.shape == (9, 9)
res_sdml = fmp.impute_features([1, 3, 5], [1, 1, 3], [0.5, 1.5, 0.5],
[1, 1, 5],
nprocs=nprocs)
assert len(fmp._res_lut) == 3
assert res_sdml[0]._x.shape == (9, 2)
assert res_sdml[1]._x.shape == (9, 2)
assert res_sdml[2]._x.shape == (9, 2)
assert not np.array_equal(res_sdml[0]._d, d)
assert not np.array_equal(res_sdml[1]._d, d)
assert not np.array_equal(res_sdml[2]._d, d)
# value
ref_res0 = np.array([[1, 1], [1, 1], [1, 2], [2, 3], [2, 5], [0, 100],
[0, 100], [0, 101], [10, 100]])
np.testing.assert_equal(res_sdml[0]._x, ref_res0)
ref_res1 = np.array([[np.median([1, 1, 2]),
np.median([1, 2, 3])],
[np.median([1, 1, 2]),
np.median([1, 2, 3])], [np.median([1, 1, 2]), 2],
[2, 3], [2, 5], [np.median([0, 0, 10]), 100],
[np.median([0, 0, 10]), 100],
[np.median([0, 0, 10]), 101], [10, 100]])
np.testing.assert_equal(res_sdml[1]._x, ref_res1)
assert res_sdml[2]._x[0, 0] > 0
assert res_sdml[2]._x[0, 1] > 0
assert res_sdml[2]._x[5, 0] > 0
assert res_sdml[2]._x[6, 0] > 0
assert res_sdml[2]._x[7, 0] > 0
# lookup
np.testing.assert_equal(fmp._res_lut[(1, 1, 0.5, 1, np.median)][0],
res_sdml[0]._x)
np.testing.assert_equal(fmp._res_lut[(3, 1, 1.5, 1, np.median)][0],
res_sdml[1]._x)
np.testing.assert_equal(fmp._res_lut[(5, 3, 0.5, 5, np.median)][0],
res_sdml[2]._x)
# d should not be changed
np.testing.assert_equal(fmp._sdm._d, d)
fmp.impute_features([1, 3, 5], [1, 1, 3], [0.5, 1.5, 0.5], [1, 1, 5],
nprocs=nprocs)
assert len(fmp._res_lut) == 3
np.testing.assert_equal(fmp._sdm._d, d)
np.testing.assert_equal(fmp._sdm.x, self.tx)
# run results should be placed with the correct order
fmp2 = knn.FeatureImputation(self.tsdm)
res1 = fmp2.impute_features([1, 3, 5], [1, 1, 3], [0.5, 1.5, 0.5],
[1, 1, 5],
nprocs=nprocs)
fmp3 = knn.FeatureImputation(self.tsdm)
res2 = fmp3.impute_features([2, 3, 4], [1, 1, 3], [0.5, 1.5, 0.5],
[1, 1, 5],
nprocs=nprocs)
res3 = fmp2.impute_features([2, 3, 4], [1, 1, 3], [0.5, 1.5, 0.5],
[1, 1, 5],
nprocs=nprocs)
np.testing.assert_equal(res1[1]._x, res3[1]._x)
np.testing.assert_equal(res2[0]._x, res3[0]._x)
np.testing.assert_equal(res2[2]._x, res3[2]._x)
