def test_graph(self):
"""
Louvain accepts graphs too.
:return:
"""
graph = matrix_to_knn_graph(self.iris.X, 30, "l2")
self.assertIsNotNone(graph)
self.assertEqual(networkx.Graph, type(graph), 1)
# basic clustering - get clusters
c = self.louvain(graph)
# First 20 iris belong to one cluster
self.assertEqual(np.ndarray, type(c))
self.assertEqual(len(self.iris), len(c))
self.assertEqual(1, len(set(c[:20].ravel())))
# clustering - get model
c = self.louvain.get_model(graph)
# First 20 iris belong to one cluster
self.assertEqual(ClusteringModel, type(c))
self.assertEqual(len(self.iris), len(c.labels))
def __init__(self,
gene_names: list,
data: np.ndarray,
neighbours: int = 30):
"""
Prepare graph for Louvain clustering. Graph construction can be perfomed in different ways, as described below.
If orange_graph=False use 1-distances as similarity for graph weights, else use Jaccard index.
:param distance_matrix: Cosine Distances between samples, square matrix
:param gene_names: List of names specifying what each row/column in distance matrix represents
:param data: Preprocessed data
:param orange_graph: Should graph be constructed with orange graph constructor
:param trimm: When orange graph is not used, if closest is None the graph will be constructed based on all
edges that would have 1-distance (weight) at least as big as trimm
:param closest: If orange_graph=False use that many closest neighbours from similarity transformed distance
matrix to construct the graph (weighs are similarities), if orange_graph=True use that many neighbours to
compute Jaccard index between nodes, which is used as a weight
"""
super().__init__(distance_matrix=None,
gene_names=gene_names,
data=data)
self._graph = orange_louvain_graph.matrix_to_knn_graph(
data=data, k_neighbors=neighbours, metric='cosine')
def run_on_data(data, normalize, pca_components, k_neighbors, metric,
resolution, state):
# type: (Table, Optional[int], int, str, float, bool, TaskState) -> Results
"""
Run the louvain clustering on `data`.
state is used to report progress and partial results. Returns early if
`task.is_interuption_requested()` returns true.
Parameters
----------
data : Table
Data table
normalize : bool
If `True`, the data is first normalized before computing PCA.
pca_components : Optional[int]
If not `None` then the data is first projected onto first
`pca_components` principal components.
k_neighbors : int
Passed to `table_to_knn_graph`
metric : str
Passed to `table_to_knn_graph`
resolution : float
Passed to `Louvain`
state : TaskState
Returns
-------
res : Results
"""
state = state # type: TaskState
res = Results(
normalize=normalize,
pca_components=pca_components,
k_neighbors=k_neighbors,
metric=metric,
resolution=resolution,
)
step = 0
if state.is_interuption_requested():
return res
if pca_components is not None:
steps = 3
state.set_status("Computing PCA...")
pca = PCA(n_components=pca_components, random_state=0)
data = res.pca_projection = pca(data)(data)
assert isinstance(data, Table)
state.set_partial_results(("pca_projection", res.pca_projection))
step += 1
else:
steps = 2
if state.is_interuption_requested():
return res
state.set_progress_value(100. * step / steps)
state.set_status("Building graph...")
# Apply Louvain preprocessing before converting the table into a graph
louvain = Louvain(resolution=resolution, random_state=0)
data = louvain.preprocess(data)
if state.is_interuption_requested():
return res
def pcallback(val):
state.set_progress_value((100. * step + 100 * val) / steps)
if state.is_interuption_requested():
raise InteruptRequested()
try:
res.graph = graph = matrix_to_knn_graph(data.X,
k_neighbors=k_neighbors,
metric=metric,
progress_callback=pcallback)
except InteruptRequested:
return res
state.set_partial_results(("graph", res.graph))
step += 1
state.set_progress_value(100 * step / steps)
state.set_status("Detecting communities...")
if state.is_interuption_requested():
return res
res.partition = louvain(graph)
state.set_partial_results(("partition", res.partition))
return res
