def test_pipeline(self):
rnds = np.random.RandomState(0)
R12 = rnds.rand(50, 30)
R13 = rnds.rand(50, 40)
R23 = rnds.rand(30, 40)
t1 = ObjectType('type1', 30)
t2 = ObjectType('type2', 40)
t3 = ObjectType('type3', 40)
relations = [
Relation(R12, t1, t2),
Relation(R13, t1, t3),
Relation(R23, t2, t3)
]
fusion_graph = FusionGraph()
fusion_graph.add_relations_from(relations)
fuser = Dfmf(random_state=rnds).fuse(fusion_graph)
self.assertEqual(fuser.factor(t1).shape, (50, 30))
self.assertEqual(fuser.factor(t2).shape, (30, 40))
self.assertEqual(fuser.factor(t3).shape, (40, 40))
self.assertEqual(fuser.backbone(relations[0]).shape, (30, 40))
self.assertEqual(fuser.backbone(relations[1]).shape, (30, 40))
self.assertEqual(fuser.backbone(relations[2]).shape, (40, 40))
new_R12 = rnds.rand(15, 30)
new_R13 = rnds.rand(15, 40)
new_relations = [Relation(new_R12, t1, t2), Relation(new_R13, t1, t3)]
new_graph = FusionGraph(new_relations)
transformer = DfmfTransform(random_state=rnds).transform(
t1, new_graph, fuser)
self.assertEqual(transformer.factor(t1).shape, (15, 30))
def test_dfmf(self):
rnds = np.random.RandomState(0)
R12 = rnds.rand(30, 30)
R13 = rnds.rand(30, 30)
t1 = ObjectType('type1', 50)
t2 = ObjectType('type2', 30)
t3 = ObjectType('type3', 10)
fusion_graph = FusionGraph()
relations = [Relation(R12, t1, t2), Relation(R13, t1, t3)]
fusion_graph.add_relations_from(relations)
fuser = Dfmf(init_type='random', random_state=rnds,
n_run=3).fuse(fusion_graph)
self.assertEqual(len(list(fuser.factor(t1))), 3)
self.assertEqual(len(list(fuser.factor(t2))), 3)
self.assertEqual(len(list(fuser.factor(t3))), 3)
self.assertEqual(len(list(fuser.backbone(relations[0]))), 3)
self.assertEqual(len(list(fuser.backbone(relations[1]))), 3)
for object_type in [t1, t2, t3]:
for factor in fuser.factor(object_type):
self.assertEqual(factor.shape, (30, object_type.rank))
G1 = fuser.factor(t1, run=1)
S13 = fuser.backbone(relations[1], run=1)
G3 = fuser.factor(t3, run=1)
R13_hat = np.dot(G1, np.dot(S13, G3.T))
completed = fuser.complete(relations[1], run=1)
np.testing.assert_almost_equal(completed, R13_hat)
def setUp(self):
rnds = np.random.RandomState(0)
X = rnds.rand(30, 30)
self.t1 = ObjectType('Type 1', 10)
self.t2 = ObjectType('Type 2', 10)
self.t3 = ObjectType('Type 3', 10)
self.t4 = ObjectType('Type 4', 10)
self.t5 = ObjectType('Type 5', 10)
self.relations1 = [
Relation(X, self.t1, self.t2, name='Test1'),
Relation(X, self.t2, self.t3),
Relation(X, self.t3, self.t4),
Relation(X, self.t4, self.t5),
Relation(X, self.t3, self.t5),
Relation(X, self.t5, self.t1)
]
self.relations2 = [
Relation(X, self.t1, self.t2, name='Test2'),
Relation(X, self.t1, self.t2),
Relation(X, self.t2, self.t3),
Relation(X, self.t3, self.t4),
Relation(X, self.t4, self.t5),
Relation(X, self.t3, self.t5),
Relation(X, self.t5, self.t1),
Relation(X, self.t4, self.t4),
Relation(X, self.t4, self.t4, name='Test3'),
Relation(X, self.t5, self.t5)
]
def load_dicty():
"""Construct fusion graph from molecular biology of Dictyostelium."""
gene = ObjectType('Gene', 50)
go_term = ObjectType('GO term', 15)
exprc = ObjectType('Experimental condition', 5)
data, rn, cn = load_source(join('dicty', 'dicty.gene_annnotations.csv.gz'))
ann = Relation(data=data,
row_type=gene,
col_type=go_term,
name='ann',
row_names=rn,
col_names=cn)
data, rn, cn = load_source(join('dicty', 'dicty.gene_expression.csv.gz'))
expr = Relation(data=data,
row_type=gene,
col_type=exprc,
name='expr',
row_names=rn,
col_names=cn)
expr.data = np.log(np.maximum(expr.data, np.finfo(np.float).eps))
data, rn, cn = load_source(join('dicty', 'dicty.ppi.csv.gz'))
ppi = Relation(data=data,
row_type=gene,
col_type=gene,
name='ppi',
row_names=rn,
col_names=cn)
return FusionGraph([ann, expr, ppi])
def test_dfmf(self):
rnds = np.random.RandomState(0)
R12_1 = np.random.rand(30, 30)
R12_2 = np.random.rand(30, 30)
R13 = np.random.rand(30, 20)
t1 = ObjectType('type1', 30)
t2 = ObjectType('type2', 30)
t3 = ObjectType('type3', 20)
relations = [
Relation(R12_1, t1, t2),
Relation(R12_2, t1, t2),
Relation(R13, t1, t3)
]
fusion_graph = FusionGraph()
fusion_graph.add_relations_from(relations)
self.assertEqual(len(fusion_graph.relations), 3)
self.assertEqual(len(fusion_graph.object_types), 3)
fuser = Dfmf(init_type='random', random_state=rnds).fuse(fusion_graph)
self.assertEqual(fuser.backbone(relations[0]).shape, (30, 30))
self.assertEqual(fuser.backbone(relations[1]).shape, (30, 30))
self.assertEqual(fuser.backbone(relations[2]).shape, (30, 20))
G1 = fuser.factor(t1)
G2 = fuser.factor(t2)
S12_1 = fuser.backbone(relations[0])
S12_2 = fuser.backbone(relations[1])
R12_1_hat = np.dot(G1, np.dot(S12_1, G2.T))
R12_2_hat = np.dot(G1, np.dot(S12_2, G2.T))
np.testing.assert_almost_equal(fuser.complete(relations[0]), R12_1_hat)
np.testing.assert_almost_equal(fuser.complete(relations[1]), R12_2_hat)
def test_transformation(self):
R12 = np.random.rand(5, 3)
t1 = ObjectType('type1', 2)
t2 = ObjectType('type2', 2)
relation = Relation(R12, t1, t2)
fusion_graph = FusionGraph()
fusion_graph.add_relation(relation)
rnds = np.random.RandomState(0)
fuser = Dfmf(init_type='random', random_state=rnds,
max_iter=100).fuse(fusion_graph)
new_R12 = R12[:2].copy()
new_graph = FusionGraph([Relation(new_R12, t1, t2)])
new_rnds = np.random.RandomState(0)
transformer = DfmfTransform(random_state=new_rnds).transform(
t1, new_graph, fuser)
new_G1 = transformer.factor(t1)
G1 = fuser.factor(t1)
G2 = fuser.factor(t2)
S12 = fuser.backbone(relation)
new_R12_hat = np.dot(new_G1, np.dot(S12, G2.T))
R12_hat = np.dot(G1, np.dot(S12, G2.T))
diff_G1 = new_G1 - G1[:2]
diff_hat = new_R12_hat - R12_hat[:2]
self.assertLess(np.sum(diff_G1**2) / diff_G1.size, 1e-5)
self.assertLess(np.sum(diff_hat**2) / diff_hat.size, 1e-5)
def test_dfmc(self):
rnds = np.random.RandomState(0)
R12 = rnds.rand(50, 30)
t1 = ObjectType('type1', 50)
t2 = ObjectType('type2', 30)
relation = Relation(R12, t1, t2)
fusion_graph = FusionGraph()
fusion_graph.add_relation(relation)
fuser = Dfmc(init_type='random', random_state=rnds).fuse(fusion_graph)
self.assertEqual(fuser.backbone(relation).shape, (50, 30))
self.assertEqual(fuser.factor(t1).shape, (50, 50))
self.assertEqual(fuser.factor(t2).shape, (30, 30))
np.testing.assert_almost_equal(fuser.complete(relation), relation.data)
def test_preprocessors(self):
rnds = np.random.RandomState(0)
R12 = rnds.rand(50, 30)
t1 = ObjectType('type1', 50)
t2 = ObjectType('type2', 30)
def preprocessor(data):
return np.ones_like(data)
relation = Relation(R12, t1, t2, preprocessor=preprocessor)
fusion_graph = FusionGraph()
fusion_graph.add_relation(relation)
fuser = Dfmf(init_type='random', random_state=rnds).fuse(fusion_graph)
self.assertEqual(fuser.backbone(relation).shape, (50, 30))
self.assertEqual(fuser.factor(t1).shape, (50, 50))
self.assertEqual(fuser.factor(t2).shape, (30, 30))
trnf = np.ones_like(relation.data)
np.testing.assert_almost_equal(fuser.complete(relation), trnf)
def test_postprocessors(self):
rnds = np.random.RandomState(0)
R12 = rnds.rand(50, 30)
R12 = np.ma.masked_greater(R12, 0.7)
t1 = ObjectType('type1', 50)
t2 = ObjectType('type2', 30)
def postprocessor(data):
return data - 10
relation = Relation(R12, t1, t2, name='R', postprocessor=postprocessor)
fusion_graph = FusionGraph()
fusion_graph.add_relation(relation)
fuser = Dfmc(init_type='random', random_state=rnds).fuse(fusion_graph)
self.assertEqual(fuser.backbone(relation).shape, (50, 30))
self.assertEqual(fuser.factor(t1).shape, (50, 50))
self.assertEqual(fuser.factor(t2).shape, (30, 30))
trnf = relation.data - 10
np.testing.assert_almost_equal(fuser.complete(relation), trnf)
np.testing.assert_equal(fusion_graph.get_relation('R').data, R12)
def run_parallel(n_jobs=1):
start_t = time.time()
n1, n2 = 500, 500
# poizkusil tudi z:
# n1, n2 = 10000, 10000
R12 = np.random.rand(n1, n2)
print(f"Number of jobs is {n_jobs}")
r1, r2 = 10, 10
# Poizkusil tudi z
# r1, r2 = 150, 150
t1 = ObjectType('type1', r1)
t2 = ObjectType('type2', r2)
relations = [Relation(R12, t1, t2)]
fusion_graph = FusionGraph(relations)
fuser = Dfmf(init_type='random_vcol', n_jobs=n_jobs).fuse(fusion_graph)
preds = fuser.complete(relations[0])
# print(f"Error is {rmse(R12, preds)}")
print(f"Done in {time.time() - start_t} sec.")
def test_infinite(self):
rnds = np.random.RandomState(0)
R12 = rnds.rand(50, 30)
R13 = rnds.rand(50, 10)
R12 = np.ma.masked_greater(R12, 0.7)
R12[R12 < 0.1] = np.nan
R13[R13 < 0.5] = np.inf
t1 = ObjectType('type1', 50)
t2 = ObjectType('type2', 30)
t3 = ObjectType('type3', 10)
relations = [
Relation(R12, t1, t2, fill_value='row_mean'),
Relation(R13, t1, t3, fill_value='col_mean')
]
fusion_graph = FusionGraph(relations)
fuser = Dfmf(init_type='random', random_state=rnds).fuse(fusion_graph)
self.assertEqual(fuser.backbone(relations[0]).shape, (50, 30))
self.assertEqual(fuser.backbone(relations[1]).shape, (50, 10))
self.assertEqual(fuser.factor(t1).shape, (50, 50))
self.assertEqual(fuser.factor(t2).shape, (30, 30))
size = np.sum(np.isfinite(fuser.complete(relations[0])))
np.testing.assert_equal(size, R12.size)
def load_pharma():
"""Construct fusion graph from the pharmacology domain."""
action = ObjectType('Action', 5)
pmid = ObjectType('PMID', 5)
depositor = ObjectType('Depositor', 5)
fingerprint = ObjectType('Fingerprint', 20)
depo_cat = ObjectType('Depositor category', 5)
chemical = ObjectType('Chemical', 10)
data, rn, cn = load_source(join('pharma', 'pharma.actions.csv.gz'))
actions = Relation(data=data,
row_type=chemical,
col_type=action,
row_names=rn,
col_names=cn)
data, rn, cn = load_source(join('pharma', 'pharma.pubmed.csv.gz'))
pubmed = Relation(data=data,
row_type=chemical,
col_type=pmid,
row_names=rn,
col_names=cn)
data, rn, cn = load_source(join('pharma', 'pharma.depositors.csv.gz'))
depositors = Relation(data=data,
row_type=chemical,
col_type=depositor,
row_names=rn,
col_names=cn)
data, rn, cn = load_source(join('pharma', 'pharma.fingerprints.csv.gz'))
fingerprints = Relation(data=data,
row_type=chemical,
col_type=fingerprint,
row_names=rn,
col_names=cn)
data, rn, cn = load_source(join('pharma', 'pharma.depo_cats.csv.gz'))
depo_cats = Relation(data=data,
row_type=depositor,
col_type=depo_cat,
row_names=rn,
col_names=cn)
data, rn, cn = load_source(join('pharma', 'pharma.tanimoto.csv.gz'))
tanimoto = Relation(data=data,
row_type=chemical,
col_type=chemical,
row_names=rn,
col_names=cn)
return FusionGraph(
[actions, pubmed, depositors, fingerprints, depo_cats, tanimoto])