def clustering_api(engines, api=None, optionables=None, prefix="clustering"):
def clustering_engine(optionables):
""" Return a default engine over a lexical graph
"""
# setup
engine = Engine("gbuilder", "clustering", "labelling")
engine.gbuilder.setup(in_name="request", out_name="graph", hidden=True)
engine.clustering.setup(in_name="graph", out_name="clusters")
engine.labelling.setup(in_name="clusters", out_name="clusters", hidden=True)
engine.gbuilder.set(engines.edge_subgraph)
engine.clustering.set(*optionables)
## Labelling
from cello.clustering.labelling.model import Label
from cello.clustering.labelling.basic import VertexAsLabel, TypeFalseLabel, normalize_score_max
def _labelling(graph, cluster, vtx):
return Label(vtx["uuid"], score=1, role="default")
labelling = VertexAsLabel( _labelling ) | normalize_score_max
engine.labelling.set(labelling)
return engine
if api is None:
api = ReliureAPI(name,expose_route = False)
## Clustering
from cello.graphs.transform import EdgeAttr
from cello.clustering.common import Infomap, Walktrap
# weighted
walktrap = Walktrap(weighted=True)
walktrap.name = "Walktrap"
infomap = Infomap(weighted=True)
infomap.name = "Infomap"
DEFAULTS = [walktrap, infomap]
if optionables == None : optionables = DEFAULTS
view = EngineView(clustering_engine(optionables))
view.set_input_type(EdgeList())
view.add_output("clusters", export_clustering, vertex_id_attr='uuid')
api.register_view(view, url_prefix=prefix)
return api
def clustering_api(graphdb,
engines,
api=None,
optionables=None,
prefix="clustering"):
def clustering_engine(optionables):
""" Return a default engine over a lexical graph
"""
# setup
engine = Engine("gbuilder", "clustering")
engine.gbuilder.setup(in_name="request", out_name="graph", hidden=True)
engine.clustering.setup(in_name="graph", out_name="clusters")
engine.gbuilder.set(engines.edge_subgraph)
engine.clustering.set(*optionables)
return engine
if api is None:
api = ReliureAPI(name, expose_route=False)
## Clustering
from cello.graphs.transform import EdgeAttr
from cello.clustering.common import Infomap, Walktrap
# weighted
walktrap = Walktrap(weighted=True)
walktrap.name = "Walktrap"
infomap = Infomap(weighted=True)
infomap.name = "Infomap"
DEFAULTS = [walktrap, infomap]
if optionables == None: optionables = DEFAULTS
from pdgapi.explor import EdgeList
view = EngineView(clustering_engine(optionables))
view.set_input_type(EdgeList())
view.add_output("clusters", export_clustering, vertex_id_attr='uuid')
api.register_view(view, url_prefix=prefix)
# cluster labels
view = EngineView(clusters_labels_engine(graphdb))
view.set_input_type(Clusters())
view.add_output("labels", lambda e: e)
api.register_view(view, url_prefix="labels")
return api
def lexical_bigraph_engine(graph):
""" Return a default engine over a lexical *bipartite* graph
"""
# setup
engine = Engine()
engine.requires("search", "clustering", "labelling", "layout")
engine.search.setup(in_name="query", out_name="graph")
engine.clustering.setup(in_name="graph", out_name="clusters")
engine.labelling.setup(in_name="clusters",
out_name="clusters",
hidden=True)
engine.layout.setup(in_name="graph", out_name="layout")
## Search
from cello.graphs.extraction import VtxMatch, VertexIds
from cello.graphs.extraction import ProxMarkovExtractionGlobalBigraph
from cello.graphs.builder import Subgraph
#HACK remove the "id" attribute (if any), it enter in conflict when exporting subgraphs to client
if 'id' in graph.vs.attributes():
del graph.vs['id']
graph_search = VtxMatch(graph, attr_list=[u"label"], default_attr=u"label")
graph_search |= ProxMarkovExtractionGlobalBigraph(graph)
graph_search |= Subgraph(graph, score_attr="prox")
graph_search.name = "ProxSearch"
#TODO: add better color to vtx
from cello.graphs.transform import VtxAttr
graph_search |= VtxAttr(color=[
(45, 200, 34),
])
#graph_search.change_option_default("vcount", 50)
engine.search.set(graph_search)
## Clustering
from cello.graphs.transform import EdgeAttr
from cello.clustering.common import Infomap, Walktrap
#RMQ infomap veux un pds, donc on en ajoute un bidon
walktrap = EdgeAttr(weight=1.) | Walktrap()
infomap = EdgeAttr(weight=1.) | Infomap()
engine.clustering.set(walktrap, infomap)
## Labelling
from cello.clustering.labelling.model import Label
from cello.clustering.labelling.basic import VertexAsLabel
engine.labelling.set(
VertexAsLabel(
lambda graph, cluster, vtx: Label(vtx["label"], role="default")))
## Layout
from cello.layout.simple import KamadaKawaiLayout
from cello.layout.proxlayout import ProxBigraphLayoutPCA
from cello.layout.transform import Shaker
engine.layout.set(
ProxBigraphLayoutPCA(dim=3) | Shaker(),
KamadaKawaiLayout(dim=3),
)
return engine
def bipartite_clustering_methods():
""" Returns standart clustering method for bipartite graphs
>>> methods = bipartite_clustering_methods()
>>> len(methods)
1
"""
methods = []
methods.append(Infomap())
return methods
def unipartite_clustering_methods():
""" Returns standart clustering method for unipartite graphs
>>> methods = unipartite_clustering_methods()
>>> len(methods)
2
"""
methods = []
methods.append(Infomap())
methods.append(Walktrap())
return methods
def lexical_graph_engine(graph):
""" Return a default engine over a lexical graph
"""
# setup
engine = Engine()
engine.requires("search", "clustering", "labelling", "layout")
engine.search.setup(in_name="query", out_name="graph")
engine.clustering.setup(in_name="graph", out_name="clusters")
engine.labelling.setup(in_name="clusters", out_name="clusters", hidden=True)
engine.layout.setup(in_name="graph", out_name="layout")
## Search
from cello.graphs.extraction import VtxMatch, ProxMarkovExtractionGlobal, VertexIds
from cello.graphs.builder import Subgraph
#HACK remove the "id" attribute (if any), it enter in conflict when exporting subgraphs to client
if 'id' in graph.vs.attributes():
del graph.vs['id']
# Pour avoir une recherche "_all"
# http://localhost:5000/verb/q/_all
@Composable
def nothing_if_all(query):
""" Make the query be nothing if it is '_all'
"""
if query in ("_all", None):
return "" #Note: p0 to [] make start from all vertices
return query
# real match search
match = VtxMatch(graph, attr_list=[u"label"], default_attr=u"label")
graph_search = nothing_if_all | match
graph_search |= ProxMarkovExtractionGlobal(graph)
graph_search |= Subgraph(graph, score_attr="prox", gdeg_attr="gdeg")
graph_search |= ProxColors(graph, graph['vertices_color'], length=5)
graph_search.name = "ProxSearch"
graph_search.change_option_default("vcount", 50)
engine.search.set(graph_search)
## Clustering
from cello.graphs.transform import EdgeAttr
from cello.clustering.common import Infomap, Walktrap
#RMQ infomap veux un pds, donc on en ajoute un bidon
walktrap = EdgeAttr(weight=1.) |Walktrap()
infomap = EdgeAttr(weight=1.) | Infomap()
engine.clustering.set(walktrap, infomap)
## Labelling
from cello.clustering.labelling.model import Label
from cello.clustering.labelling.basic import VertexAsLabel
vertex_from_vtx = lambda graph, cluster, vtx: Label(vtx["label"], role="default", score=vtx["gdeg"])
engine.labelling.set(VertexAsLabel(vertex_from_vtx))
## Layout
from cello.layout.simple import KamadaKawaiLayout
from cello.layout.proxlayout import ProxLayoutRandomProj
from cello.layout.proxlayout import ProxLayoutPCA
from cello.layout.transform import Shaker
engine.layout.set(
ProxLayoutPCA(dim=3) | Shaker(),
KamadaKawaiLayout(dim=3),
)
return engine
def engine(index):
""" Return a default engine over a lexical graph
"""
# setup
from reliure.engine import Engine
engine = Engine("graph", "clustering", "labelling", "layout")
engine.graph.setup(in_name="query", out_name="graph")
engine.clustering.setup(in_name="graph", out_name="clusters")
engine.labelling.setup(in_name="clusters",
out_name="clusters",
hidden=True)
engine.layout.setup(in_name="graph", out_name="layout")
## Search
def tmuse_subgraph(query, length=50):
return tmuse.subgraph(index, query, length=length)
graph_search = Optionable("GraphSearch")
graph_search._func = Composable(tmuse_subgraph)
graph_search.add_option("length", Numeric(vtype=int, default=50))
from cello.graphs.transform import VtxAttr
graph_search |= VtxAttr(color=[
(45, 200, 34),
])
graph_search |= VtxAttr(type=1)
engine.graph.set(graph_search)
## Clustering
from cello.graphs.transform import EdgeAttr
from cello.clustering.common import Infomap, Walktrap
#RMQ infomap veux un pds, donc on en ajoute un bidon
walktrap = EdgeAttr(weight=1.) | Walktrap()
infomap = EdgeAttr(weight=1.) | Infomap()
engine.clustering.set(infomap, walktrap)
## Labelling
from cello.clustering.labelling.model import Label
from cello.clustering.labelling.basic import VertexAsLabel, TypeFalseLabel, normalize_score_max
def _labelling(graph, cluster, vtx):
score = TypeFalseLabel.scoring_prop_ofclust(graph, cluster, vtx)
return Label(vtx["form"], score=score, role="default")
labelling = VertexAsLabel(_labelling) | normalize_score_max
engine.labelling.set(labelling)
## Layout
from cello.layout.simple import KamadaKawaiLayout
#from cello.layout.proxlayout import ProxLayoutRandomProj
from cello.layout.proxlayout import ProxLayoutPCA
from cello.layout.transform import Shaker
from cello.layout.transform import ByConnectedComponent, normalise
default_layout = ProxLayoutPCA(dim=3,
name="ProxPca3d") | Shaker(kelastic=.9)
default_layout = KamadaKawaiLayout(
dim=3, name="KamadaKawaiLayout") | Shaker(kelastic=.9)
engine.layout.set(
ByConnectedComponent(default_layout) | normalise,
#KamadaKawaiLayout(dim=3, name="KamadaKawai3D"),
#ProxLayoutPCA(dim=2, name="ProxPca2d") | Shaker(kelastic=1.8),
#KamadaKawaiLayout(dim=2, name="KamadaKawai2D")
)
return engine