def test_transform_sparse_adaptive_pca():
G = build_graph(data, sparse=True, n_pca=True, random_state=42)
assert np.all(G.data_nu == G.transform(G.data))
with assert_raises_message(
ValueError,
"data of shape ({0}, 1) cannot be transformed to graph built on data of shape ({0}, {1}). Expected shape ({0}, {1})".format(
G.data.shape[0], G.data.shape[1]
),
):
G.transform(sp.csr_matrix(G.data)[:, 0])
with assert_raises_message(
ValueError,
"data of shape ({0}, 15) cannot be transformed to graph built on data of shape ({0}, {1}). Expected shape ({0}, {1})".format(
G.data.shape[0], G.data.shape[1]
),
):
G.transform(sp.csr_matrix(G.data)[:, :15])
G2 = build_graph(
data, sparse=True, n_pca=True, rank_threshold=G.rank_threshold, random_state=42
)
assert np.allclose(G2.data_nu, G2.transform(G2.data))
assert np.allclose(G2.data_nu, G.transform(G.data))
G3 = build_graph(data, sparse=True, n_pca=G2.n_pca, random_state=42)
assert np.allclose(G3.data_nu, G3.transform(G3.data))
assert np.allclose(G3.data_nu, G2.transform(G2.data))
def test_check_between():
graphtools.utils.check_between(-5, -3, foo=-4)
with assert_raises_message(ValueError,
"Expected foo between -5 and -3, got -6"):
graphtools.utils.check_between(-5, -3, foo=-6)
with assert_raises_message(ValueError, "Expected v_max > -3, got -5"):
graphtools.utils.check_between(-3, -5, foo=-6)
def test_inverse_transform_dense_no_pca():
G = build_graph(data, n_pca=None)
np.testing.assert_allclose(
data[:, 5:7], G.inverse_transform(G.data_nu, columns=[5, 6]), atol=1e-12
)
assert np.all(G.data == G.inverse_transform(G.data_nu))
with assert_raises_message(
ValueError,
"data of shape ({0},) cannot be inverse transformed from graph built on reduced data of shape ({0}, {1})".format(
data.shape[0], G.data.shape[1]
),
):
G.inverse_transform(G.data[:, 0])
with assert_raises_message(
ValueError,
"data of shape ({0}, 1, 15) cannot be inverse transformed from graph built on reduced data of shape ({0}, {1})".format(
data.shape[0], data.shape[1]
),
):
G.inverse_transform(G.data[:, None, :15])
with assert_raises_message(
ValueError,
"data of shape ({0}, 15) cannot be inverse transformed from graph built on reduced data of shape ({0}, {1})".format(
data.shape[0], data.shape[1]
),
):
G.inverse_transform(G.data[:, :15])
def test_inverse_transform_dense_pca():
G = build_graph(data, n_pca=data.shape[1] - 1)
np.testing.assert_allclose(G.data, G.inverse_transform(G.data_nu), atol=1e-12)
np.testing.assert_allclose(
G.data[:, -1, None], G.inverse_transform(G.data_nu, columns=-1), atol=1e-12
)
np.testing.assert_allclose(
G.data[:, 5:7], G.inverse_transform(G.data_nu, columns=[5, 6]), atol=1e-12
)
with assert_raises_message(
IndexError,
"index {0} is out of bounds for axis 1 with size {0}".format(G.data.shape[1]),
):
G.inverse_transform(G.data_nu, columns=data.shape[1])
with assert_raises_message(
ValueError,
"data of shape ({0},) cannot be inverse transformed from graph built on reduced data of shape ({0}, {1})".format(
G.data.shape[0], G.n_pca
),
):
G.inverse_transform(G.data[:, 0])
with assert_raises_message(
ValueError,
"data of shape ({0}, 1, 15) cannot be inverse transformed from graph built on reduced data of shape ({0}, {1})".format(
G.data.shape[0], G.n_pca
),
):
G.inverse_transform(G.data[:, None, :15])
with assert_raises_message(
ValueError,
"data of shape ({0}, 15) cannot be inverse transformed from graph built on reduced data of shape ({0}, {1})".format(
G.data.shape[0], G.n_pca
),
):
G.inverse_transform(G.data[:, :15])
def test_set_params():
X, sample_idx = generate_swiss_roll()
G = build_graph(X,
sample_idx=sample_idx,
kernel_symm="mnn",
theta=0.5,
n_pca=None,
thresh=1e-4)
assert G.get_params() == {
"n_pca": None,
"random_state": 42,
"kernel_symm": "mnn",
"theta": 0.5,
"anisotropy": 0,
"beta": 1,
"knn": 3,
"decay": 10,
"bandwidth": None,
"distance": "euclidean",
"thresh": 1e-4,
"n_jobs": 1,
}
G.set_params(n_jobs=4)
assert G.n_jobs == 4
for graph in G.subgraphs:
assert graph.n_jobs == 4
assert graph.knn_tree.n_jobs == 4
G.set_params(random_state=13)
assert G.random_state == 13
for graph in G.subgraphs:
assert graph.random_state == 13
G.set_params(verbose=2)
assert G.verbose == 2
for graph in G.subgraphs:
assert graph.verbose == 2
G.set_params(verbose=0)
with assert_raises_message(ValueError,
"Cannot update knn. Please create a new graph"):
G.set_params(knn=15)
with assert_raises_message(
ValueError, "Cannot update decay. Please create a new graph"):
G.set_params(decay=15)
with assert_raises_message(
ValueError, "Cannot update distance. Please create a new graph"):
G.set_params(distance="manhattan")
with assert_raises_message(
ValueError, "Cannot update thresh. Please create a new graph"):
G.set_params(thresh=1e-3)
with assert_raises_message(
ValueError, "Cannot update beta. Please create a new graph"):
G.set_params(beta=0.2)
G.set_params(knn=G.knn,
decay=G.decay,
thresh=G.thresh,
distance=G.distance,
beta=G.beta)
def test_1d_data():
with assert_raises_message(
ValueError,
"Expected 2D array, got 1D array instead (shape: ({},).)".format(data.shape[0]),
):
build_graph(data[:, 0])
with assert_raises_message(
ValueError,
"Reshape your data either using array.reshape(-1, 1) "
"if your data has a single feature or array.reshape(1, -1) if "
"it contains a single sample.".format(data.shape[0]),
):
build_graph(data[:, 0])
def test_sample_idx_wrong_length():
with assert_raises_message(
ValueError,
"sample_idx (10) must be the same length as data ({})".format(
data.shape[0]),
):
build_graph(data, graphtype="mnn", sample_idx=np.arange(10))
def test_precomputed_interpolate():
with assert_raises_message(ValueError,
"Cannot extend kernel on precomputed graph"):
G = build_graph(squareform(pdist(data)),
n_pca=None,
precomputed="distance")
G.build_kernel_to_data(data)
def test_precomputed_invalid():
with assert_raises_message(
ValueError,
"Precomputed value invalid not recognized. Choose from ['distance', 'affinity', 'adjacency']",
):
build_graph(np.random.uniform(0, 1, [200, 200]),
precomputed="invalid",
n_pca=None)
def test_transform_sparse_pca():
G = build_graph(data, sparse=True, n_pca=20)
assert np.all(G.data_nu == G.transform(G.data))
with assert_raises_message(
ValueError,
"data of shape ({0}, 1) cannot be transformed to graph built on data of shape ({0}, {1}). Expected shape ({0}, {1})".format(
G.data.shape[0], G.data.shape[1]
),
):
G.transform(sp.csr_matrix(G.data)[:, 0])
with assert_raises_message(
ValueError,
"data of shape ({0}, 15) cannot be transformed to graph built on data of shape ({0}, {1}). Expected shape ({0}, {1})".format(
G.data.shape[0], G.data.shape[1]
),
):
G.transform(sp.csr_matrix(G.data)[:, :15])
def test_build_landmark_with_too_many_landmarks():
with assert_raises_message(
ValueError,
"n_landmark ({0}) >= n_samples ({0}). Use kNNGraph instead".format(
data.shape[0]
),
):
build_graph(data, n_landmark=len(data))
def test_shortest_path_invalid():
with assert_raises_message(
ValueError,
"Expected `distance` in ['constant', 'data', 'affinity']. Got invalid",
):
data_small = data[np.random.choice(len(data), len(data) // 4, replace=False)]
G = build_graph(data_small, knn=5, decay=None)
G.shortest_path(distance="invalid")
def test_shortest_path_no_decay_affinity():
with assert_raises_message(
ValueError,
"Graph shortest path with affinity distance only valid for weighted graphs. For unweighted graphs, use `distance='constant'` or `distance='data'`.",
):
data_small = data[np.random.choice(len(data), len(data) // 4, replace=False)]
G = build_graph(data_small, knn=5, decay=None)
G.shortest_path(distance="affinity")
def test_3d_data():
with assert_raises_message(
ValueError,
"Expected 2D array, got 3D array instead (shape: ({0}, 64, 1).)".format(
data.shape[0]
),
):
build_graph(data[:, :, None])
def test_inverse_transform_sparse_no_pca():
G = build_graph(data, sparse=True, n_pca=None)
assert np.sum(G.data != G.inverse_transform(G.data_nu)) == 0
with assert_raises_message(
ValueError,
"data of shape ({0}, 1) cannot be inverse transformed from graph built on reduced data of shape ({0}, {1})".format(
G.data.shape[0], G.data.shape[1]
),
):
G.inverse_transform(sp.csr_matrix(G.data)[:, 0])
with assert_raises_message(
ValueError,
"data of shape ({0}, 15) cannot be inverse transformed from graph built on reduced data of shape ({0}, {1})".format(
G.data.shape[0], G.data.shape[1]
),
):
G.inverse_transform(sp.csr_matrix(G.data)[:, :15])
def test_knn_no_knn_no_bandwidth():
with assert_raises_message(
ValueError, "Either `knn` or `bandwidth` must be provided."):
build_graph(data,
graphtype="knn",
knn=None,
bandwidth=None,
thresh=1e-4)
def test_transform_sparse_no_pca():
G = build_graph(data, sparse=True, n_pca=None)
assert np.sum(G.data_nu != G.transform(G.data)) == 0
with assert_raises_message(
ValueError,
"data of shape {} cannot be transformed to graph built on data of shape {}".format(
G.data.tocsr()[:, 0].shape, G.data.shape
),
):
G.transform(sp.csr_matrix(G.data)[:, 0])
with assert_raises_message(
ValueError,
"data of shape {} cannot be transformed to graph built on data of shape {}".format(
G.data.tocsr()[:, :15].shape, G.data.shape
),
):
G.transform(sp.csr_matrix(G.data)[:, :15])
def test_shortest_path_precomputed_no_decay_data():
with assert_raises_message(
ValueError,
"Graph shortest path with data distance not valid for precomputed graphs. For precomputed graphs, use `distance='constant'` for unweighted graphs and `distance='affinity'` for weighted graphs.",
):
data_small = data[np.random.choice(len(data), len(data) // 4, replace=False)]
G = build_graph(data_small, knn=5, decay=None)
G = graphtools.Graph(G.K, precomputed="affinity")
G.shortest_path(distance="data")
def test_knn_interpolate_wrong_shape():
G = build_graph(data, n_pca=10, decay=None)
with assert_raises_message(
ValueError,
"Expected a 2D matrix. Y has shape ({},)".format(data.shape[0])):
G.extend_to_data(data[:, 0])
with assert_raises_message(
ValueError,
"Expected a 2D matrix. Y has shape ({}, {}, 1)".format(
data.shape[0], data.shape[1]),
):
G.extend_to_data(data[:, :, None])
with assert_raises_message(ValueError,
"Y must be of shape either (n, 64) or (n, 10)"):
G.extend_to_data(data[:, :data.shape[1] // 2])
G = build_graph(data, n_pca=None, decay=None)
with assert_raises_message(ValueError, "Y must be of shape (n, 64)"):
G.extend_to_data(data[:, :data.shape[1] // 2])
def test_sample_idx_unique():
with assert_raises_message(
ValueError, "sample_idx must contain more than one unique value"):
build_graph(data,
graph_class=graphtools.graphs.MNNGraph,
sample_idx=np.ones(len(data)))
with assert_warns_message(UserWarning,
"Only one unique sample. Not using MNNGraph"):
build_graph(data, sample_idx=np.ones(len(data)), graphtype="mnn")
def test_build_exact_with_sample_idx():
with assert_raises_message(
ValueError,
"TraditionalGraph does not support batch correction. Use `graphtype='mnn'` or `sample_idx=None`",
):
build_graph(data,
graphtype="exact",
sample_idx=np.arange(len(data)),
decay=10)
def test_sample_idx_and_precomputed():
with assert_raises_message(
ValueError,
"MNNGraph does not support precomputed values. Use `graphtype='exact'` and `sample_idx=None` or `precomputed=None`",
):
build_graph(data,
n_pca=None,
sample_idx=np.arange(10),
precomputed="distance")
def test_knn_graph_invalid_symm():
with assert_raises_message(
ValueError,
"kernel_symm 'invalid' not recognized. Choose from '+', '*', 'mnn', or 'none'.",
):
build_graph(data,
graphtype="knn",
knn=5,
thresh=1e-4,
kernel_symm="invalid")
def test_set_params():
G = graphtools.base.Data(data, n_pca=20)
assert G.get_params() == {"n_pca": 20, "random_state": None}
G.set_params(random_state=13)
assert G.random_state == 13
with assert_raises_message(
ValueError, "Cannot update n_pca. Please create a new graph"
):
G.set_params(n_pca=10)
G.set_params(n_pca=G.n_pca)
def test_shortest_path_decay_data():
with assert_raises_message(
NotImplementedError,
"Graph shortest path with constant or data distance only implemented for unweighted graphs. For weighted graphs, use `distance='affinity'`.",
):
data_small = data[np.random.choice(len(data),
len(data) // 4,
replace=False)]
G = build_graph(data_small, knn=5, decay=15)
G.shortest_path(distance="data")
def test_exact_interpolate():
G = build_graph(data, decay=10, thresh=0)
with assert_raises_message(
ValueError, "Either `transitions` or `Y` must be provided."):
G.interpolate(data)
pca_data = PCA(2).fit_transform(data)
transitions = G.extend_to_data(data)
assert np.all(
G.interpolate(pca_data, Y=data) == G.interpolate(
pca_data, transitions=transitions))
def test_knn_interpolate():
G = build_graph(data, decay=None)
with assert_raises_message(
ValueError, "Either `transitions` or `Y` must be provided."):
G.interpolate(data)
pca_data = PCA(2).fit_transform(data)
transitions = G.extend_to_data(data)
np.testing.assert_equal(
G.interpolate(pca_data, Y=data),
G.interpolate(pca_data, transitions=transitions),
)
def test_shortest_path_precomputed_decay_constant():
with assert_raises_message(
NotImplementedError,
"Graph shortest path with constant distance only implemented for unweighted graphs. For weighted graphs, use `distance='affinity'`.",
):
data_small = data[np.random.choice(len(data),
len(data) // 4,
replace=False)]
G = build_graph(data_small, knn=5, decay=15)
G = graphtools.Graph(G.K, precomputed="affinity")
G.shortest_path(distance="constant")
def test_mnn_with_string_theta():
with assert_raises_message(
TypeError, "Expected `theta` as a float. Got <class 'str'>."):
build_graph(
data,
thresh=0,
n_pca=20,
decay=10,
knn=5,
random_state=42,
sample_idx=digits["target"],
kernel_symm="mnn",
theta="invalid",
)
def test_inverse_transform_sparse_svd():
G = build_graph(data, sparse=True, n_pca=data.shape[1] - 1)
np.testing.assert_allclose(data, G.inverse_transform(G.data_nu), atol=1e-12)
np.testing.assert_allclose(
data[:, -1, None], G.inverse_transform(G.data_nu, columns=-1), atol=1e-12
)
np.testing.assert_allclose(
data[:, 5:7], G.inverse_transform(G.data_nu, columns=[5, 6]), atol=1e-12
)
with assert_raises_message(
IndexError, "index 64 is out of bounds for axis 1 with size 64"
):
G.inverse_transform(G.data_nu, columns=data.shape[1])
with assert_raises_message(
TypeError,
"A sparse matrix was passed, but dense data is required. Use X.toarray() to convert to a dense numpy array.",
):
G.inverse_transform(sp.csr_matrix(G.data)[:, 0])
with assert_raises_message(
TypeError,
"A sparse matrix was passed, but dense data is required. Use X.toarray() to convert to a dense numpy array.",
):
G.inverse_transform(sp.csr_matrix(G.data)[:, :15])
with assert_raises_message(
ValueError,
"data of shape ({0},) cannot be inverse transformed from graph built on reduced data of shape ({0}, {1}). Expected shape ({0}, {1})".format(
data.shape[0], G.n_pca
),
):
G.inverse_transform(data[:, 0])
with assert_raises_message(
ValueError,
"data of shape ({0}, 15) cannot be inverse transformed from graph built on reduced data of shape ({0}, {1}). Expected shape ({0}, {1})".format(
data.shape[0], G.n_pca
),
):
G.inverse_transform(data[:, :15])
