def gen_graph_from_mongo(self):
'''
load graph structure from storage. note that add_edge_list will not
match vertex ids (str ids) in subsequent calls of the function
'''
self.from_storage(find={'projection': {'material_id': 1, 'edges': 1}})
sources = self.memory['material_id']
destinations = self.memory['edges']
self.memory = None # cleanup memory attribute
print('loaded data structures')
edge_list = [(sources[i], destinations[i][j])
for i in range(len(sources))
for j in range(len(destinations[i]))]
print('generated edge list')
sources = None # cleanup temporary data variables
destinations = None
graph = Graph(directed=False)
graph.add_edge_list(edge_list, hashed=True, string_vals=True)
return graph
def test_feasibility(g, weights):
internal_g = from_gt(g, weights)
edges = minimum_branching(internal_g, [0])
tree = Graph(directed=True)
tree.add_edge_list(edges)
assert is_arborescence(tree)
def alignment_graph(lengths=[], pairings=[], alignments=[]):
#print('making graph')
g = Graph(directed=False)
seq_index = g.new_vertex_property("int")
time = g.new_vertex_property("int")
#add vertices
g.add_vertex(sum(lengths))
seq_index.a = np.concatenate([np.repeat(i,l) for i,l in enumerate(lengths)])
time.a = np.concatenate([np.arange(l) for l in lengths])
#add edges (alignments)
alignment_index = g.new_edge_property("int")
segment_index = g.new_edge_property("int")
for i,a in enumerate(alignments):
if len(a) > 0:
j, k = pairings[i]
pairs = np.concatenate(a, axis=0)
indicesJ = (np.arange(lengths[j]) + sum(lengths[:j]))[pairs.T[0]]
indicesK = (np.arange(lengths[k]) + sum(lengths[:k]))[pairs.T[1]]
seg_indices = np.concatenate([np.repeat(i, len(a))
for i,a in enumerate(a)])
g.add_edge_list(np.vstack([indicesJ, indicesK,
np.repeat(i, len(pairs)), seg_indices]).T,
eprops=[alignment_index, segment_index])
#g.add_edge_list([(b, a) for (a, b) in g.edges()])
#print('created alignment graph', g)
#g = prune_isolated_vertices(g)
#print('pruned alignment graph', g)
#g = transitive_closure(g)
#graph_draw(g, output_size=(1000, 1000), output="results/casey_jones_bars.pdf")
return g, seq_index, time, alignment_index, segment_index
def construct_motif_graph(graph_container, motif, vertex_maps=None):
"""Construct and return a undirected gt graph containing
motif relationship. Note that graph_tool generates empty nodes
to fill in the missing indices. For example, if we add edge (1,2)
to an empty graph, the graph will have 3 nodes: 0, 1, 2 and 1 edge (1,2).
For this reason, the returned `m_graph` usually has a large number of
disconnected nodes.
Parameters:
graph_container - GraphContainer - Store the original network
motif - Motif - Motif in study
Returns:
m_graph - gt.Graph - Undirected graph for motif cooccurence
"""
if motif.anchors is None:
print("Warning: Turning motif groups into cliques.")
graph = graph_container.get_gt_graph()
graph.set_directed(motif.gt_motif.is_directed())
# graph_tool.Graph
m_graph = Graph(directed=False)
if vertex_maps is None:
m, c, vertex_maps = count_motif(graph, motif)
for prop_list in vertex_maps:
for prop in prop_list:
edges = [i for i in motif.anchored_edges(graph, prop.get_array())]
m_graph.add_edge_list(edges)
return m_graph
def gen_sub_graph_from_mongo(self, center, snn=1):
'''
load graph structure from storage. note that add_edge_list will not
match vertex ids (str ids) in subsequent calls of the function
Args:
center (str) mp-id of the center of the graph
snn (int) the number of second nearest neighbors to expand to
'''
edge_list = []
self.from_storage(
find={
'filter': {
'material_id': center
},
'projection': {
'material_id': 1,
'edges': 1
}
})
sources = self.memory['material_id'][0]
destinations = self.memory['edges'][0]
edge_list.extend([(sources, destinations[j])
for j in range(len(destinations))])
for i in range(snn):
self.from_storage(
find={
'filter': {
'material_id': {
'$in': destinations
}
},
'projection': {
'material_id': 1,
'edges': 1
}
})
sources = self.memory['material_id']
destinations = self.memory['edges']
edge_list.extend([(sources[i], destinations[i][j])
for i in range(len(sources))
for j in range(len(destinations[i]))])
destinations = [
destinations[i][j] for i in range(len(sources))
for j in range(len(destinations[i]))
]
print('generated edge list')
graph = Graph(directed=False)
graph.add_edge_list(edge_list, hashed=True, string_vals=True)
return graph
def test_feasibility(g, weights):
edges = [(e[0], e[1], w) for e, w in zip(g.get_edges(), weights)]
min_edges = find_minimum_branching(g.num_vertices(), edges, roots=[0])
tree = Graph(directed=True)
tree.add_edge_list(min_edges)
assert is_arborescence(tree)
def graph_from_matrix(matrix, directed=False):
g = Graph(directed=directed)
g.add_vertex(len(matrix))
weights = g.new_ep("float")
edges = np.nonzero(matrix)
edges = np.append(edges, [matrix[edges]], axis=0)
g.add_edge_list(list(zip(*edges)), eprops=[weights])
#graph_draw(g, output_size=(1000, 1000), output="results/structure.pdf")
return g, weights
def test_graphtool():
g = Graph(directed=True)
g.add_vertex(4)
g.add_edge_list([(0, 1), (1, 2), (2, 3), (3, 0)])
weight = g.new_edge_property('float')
weight[g.edge(0, 1)] = 1
weight[g.edge(1, 2)] = 2
weight[g.edge(2, 3)] = 3
weight[g.edge(3, 0)] = 4
assert set(gt2edges_and_weights(g, weight)) == {
(0, 1, 1), (1, 2, 2), (2, 3, 3), (3, 0, 4)
}
class ZonedNetwork:
def __init__(self,
size: Tuple[int] = (10, 10),
field_size: Tuple[int] = (100, 100)):
self.g = Graph(directed=True)
self.n_zones = size[0] * size[1]
self.fwidth = field_size[0]
self.fheight = field_size[1]
self.n_rows = size[0]
self.n_cols = size[1]
self.row_size: float = self.fheight / self.n_rows
self.col_size: float = self.fwidth / self.n_cols
self.g.add_vertex(self.n_zones)
def get_zone(self, coords: Tuple):
r = int(coords[1] / self.row_size)
c = int(coords[0] / self.col_size)
r = min(self.n_rows - 1, r)
c = min(self.n_cols - 1, c)
return self.g.vertex(r * self.n_cols + c)
def add_passes(self, coords_pairs: List[Tuple]):
pairs = [(self.get_zone((x1, y1)), self.get_zone((x2, y2)))
for x1, y1, x2, y2 in coords_pairs]
return self.g.add_edge_list(pairs)
def save(self, file: str):
self.g.save(file, fmt='graphml')
def clean_up(g, seg_index):
#plot_matrix(np.triu(adjacency_matrix(g)), "results/clean0.png")
#graph_draw(g, output_size=(1000, 1000), output="results/clean_up0.pdf")
seg_combos = get_segment_combos(g, seg_index)
best = sorted(seg_combos.items(), key=lambda c: c[1], reverse=True)#[:200]
#print(best)
best = best[0][0]
#print(best)
#print(edges[:100])
reduced = Graph(directed=False)
reduced.add_vertex(len(g.get_vertices()))
edges = g.get_edges([seg_index])
edges = edges[np.where(np.isin(edges[:,2], best))]
reduced.add_edge_list(edges)
#print(reduced)
#plot_matrix(np.triu(adjacency_matrix(reduced)), "results/cleani2.png")
#graph_draw(reduced, output_size=(1000, 1000), output="results/clean_up1.pdf")
return reduced
def build_graph(m_codes, m_list):
n_models, n_attributes = m_codes.shape
g = Graph()
v_map = {}
names = g.new_vertex_property("object")
v_atts = g.add_vertex(n_attributes)
v_mods = g.add_vertex(n_models)
v_imps = g.add_vertex(n_attributes)
for v_idx, v in enumerate(v_atts):
v_n = v_name(v_idx, kind="data")
v_map[v_n] = int(v)
names[v] = v_n
for v_idx, v in enumerate(v_mods):
v_n = v_name(v_idx, kind="model")
v_map[v_n] = int(v)
names[v] = v_n
in_edges = ((d, v) for d in m_list[v_idx].desc_ids)
out_edges = ((v, t) for t in m_list[v_idx].targ_ids)
g.add_edge_list(in_edges)
g.add_edge_list(out_edges)
for v_idx, v in enumerate(v_imps):
v_n = v_name(v_idx, kind="imputation")
v_map[v_n] = int(v)
names[v] = v_n
g.vp.names = names
g.v_map = v_map
return g
def get_pagerank_values(self):
start = time.time()
logger.info('Started call to get_pagerank')
g = Graph()
vp = g.add_edge_list(self.__v.get_graph_edges(),
hashed=True,
hash_type='int')
logger.info('Delta time to build graph: {}s'.format(
timedelta(seconds=(time.time() - start))))
start = time.time()
ranks = pagerank(g)
logger.info('Delta time to compute pagerank: {}s'.format(
timedelta(seconds=(time.time() - start))))
for vertex in g.vertices():
qid = vp[vertex]
r = ranks[vertex]
yield qid, r
dict_map = pickle.load(handle)
print (dict_map)
print ("Carregando arquivo...")
g = Graph(directed=False)
edgelist = []
with open(args.edge_list) as f:
for line in f:
if(line):
edgelist.append(map(int,line.split()))
labels_vertices = g.add_edge_list(edgelist,hashed=True)
labels_vertices_str = g.new_vertex_property("string")
for v in g.vertices():
labels_vertices_str[v] = str(labels_vertices[v])
labels_vertices_inv = mapeiaLabels(g,labels_vertices)
pos = sfdp_layout(g)
colors = trataCores(dict_map)
pos_new = trataPosicoes(g,pos,dict_map,labels_vertices_inv)
color = g.new_vertex_property("string")
for v in g.vertices():
index = g.vertex_index[v]
class GeneralGraph():
"""
General wrapper for graph-tool or networkx graphs to add edges and nodes
according to constraints
"""
def __init__(self, directed=True, verbose=1):
self.graphtool = GRAPH_TOOL
# Initialize graph
if self.graphtool:
self.graph = Graph(directed=directed)
self.weight = self.graph.new_edge_property("float")
else:
if directed:
print("directed graph")
self.graph = nx.DiGraph()
else:
self.graph = nx.Graph()
# set metaparameter
self.time_logs = {}
self.verbose = verbose
def set_edge_costs(self,
layer_classes=["resistance"],
class_weights=[1],
**kwargs):
"""
Initialize edge cost variables
:param classes: list of cost categories
:param weights: list of weights for cost categories - must be of same
shape as classes (if None, then equal weighting)
"""
class_weights = np.array(class_weights)
# set different costs:
self.cost_classes = layer_classes
if self.graphtool:
self.cost_props = [
self.graph.new_edge_property("float")
for _ in range(len(layer_classes))
]
self.cost_weights = class_weights / np.sum(class_weights)
if self.verbose:
print(self.cost_classes, self.cost_weights)
# save weighted instance for plotting
self.instance = np.sum(
np.moveaxis(self.cost_instance, 0, -1) * self.cost_weights,
axis=2) * self.hard_constraints
def set_shift(self,
start,
dest,
pylon_dist_min=3,
pylon_dist_max=5,
max_angle=np.pi / 2,
**kwargs):
"""
Initialize shift variable by getting the donut values
:param lower, upper: min and max distance of pylons
:param vec: vector of diretion of edges
:param max_angle: Maximum angle of edges to vec
"""
vec = dest - start
if self.verbose:
print("SHIFT:", pylon_dist_min, pylon_dist_max, vec, max_angle)
self.shifts = get_half_donut(pylon_dist_min,
pylon_dist_max,
vec,
angle_max=max_angle)
self.shift_tuples = self.shifts
def set_corridor(self,
dist_surface,
start_inds,
dest_inds,
sample_func="mean",
sample_method="simple",
factor_or_n_edges=1):
# set new corridor
corridor = (dist_surface > 0).astype(int)
self.factor = factor_or_n_edges
self.cost_rest = self.cost_instance * (self.hard_constraints >
0).astype(int) * corridor
# downsample
tic = time.time()
if self.factor > 1:
self.cost_rest = CostUtils.downsample(self.cost_rest,
self.factor,
mode=sample_method,
func=sample_func)
self.time_logs["downsample"] = round(time.time() - tic, 3)
# repeat because edge artifacts
self.cost_rest = self.cost_rest * (self.hard_constraints >
0).astype(int) * corridor
# add start and end TODO ugly
self.cost_rest[:, dest_inds[0],
dest_inds[1]] = self.cost_instance[:, dest_inds[0],
dest_inds[1]]
self.cost_rest[:, start_inds[0],
start_inds[1]] = self.cost_instance[:, start_inds[0],
start_inds[1]]
def add_nodes(self, nodes):
"""
Add vertices to the graph
param nodes: list of node names if networkx, integer if graphtool
"""
tic = time.time()
# add nodes to graph
if self.graphtool:
_ = self.graph.add_vertex(nodes)
self.n_nodes = len(list(self.graph.vertices()))
else:
self.graph.add_nodes_from(np.arange(nodes))
self.n_nodes = len(self.graph.nodes())
# verbose
if self.verbose:
print("Added nodes:", nodes, "in time:", time.time() - tic)
self.time_logs["add_nodes"] = round(time.time() - tic, 3)
def add_edges(self):
tic_function = time.time()
n_edges = 0
# kernels, posneg = ConstraintUtils.get_kernel(self.shifts,
# self.shift_vals)
# edge_array = []
times_edge_list = []
times_add_edges = []
if self.verbose:
print("n_neighbors:", len(self.shift_tuples))
for i in range(len(self.shift_tuples)):
tic_edges = time.time()
# set cost rest if necessary (random graph)
self.set_cost_rest()
# compute shift and weights
out = self._compute_edges(self.shift_tuples[i])
# Error if -1 entries because graph-tool crashes with -1 nodes
if np.any(out[:, :2].flatten() < 0):
print(np.where(out[:, :2] < 0))
raise RuntimeError
n_edges += len(out)
times_edge_list.append(round(time.time() - tic_edges, 3))
# add edges to graph
tic_graph = time.time()
if self.graphtool:
self.graph.add_edge_list(out, eprops=self.cost_props)
else:
nx_edge_list = [(e[0], e[1], {
"weight": np.sum(e[2:] * self.cost_weights)
}) for e in out]
self.graph.add_edges_from(nx_edge_list)
times_add_edges.append(round(time.time() - tic_graph, 3))
# alternative: collect edges here and add alltogether
# edge_array.append(out)
# # alternative: add edges all in one go
# tic_concat = time.time()
# edge_lists_concat = np.concatenate(edge_array, axis=0)
# self.time_logs["concatenate"] = round(time.time() - tic_concat, 3)
# print("time for concatenate:", self.time_logs["concatenate"])
# tic_graph = time.time()
# self.graph.add_edge_list(edge_lists_concat, eprops=[self.weight])
# self.time_logs["add_edges"] = round(
# (time.time() - tic_graph) / len(shifts), 3
# )
self.n_edges = len(list(self.graph.edges()))
self._update_time_logs(times_add_edges, times_edge_list, tic_function)
if self.verbose:
print("DONE adding", n_edges, "edges:", time.time() - tic_function)
def _update_time_logs(self, times_add_edges, times_edge_list,
tic_function):
self.time_logs["add_edges"] = round(np.mean(times_add_edges), 3)
self.time_logs["add_edges_times"] = times_add_edges
self.time_logs["edge_list"] = round(np.mean(times_edge_list), 3)
self.time_logs["edge_list_times"] = times_edge_list
self.time_logs["add_all_edges"] = round(time.time() - tic_function, 3)
if self.verbose:
print("Done adding edges:", len(list(self.graph.edges())))
def sum_costs(self):
"""
Additive weighting of costs
Take the individual edge costs, compute weighted sum --> self.weight
"""
# add sum of all costs
if not self.graphtool:
return
tic = time.time()
summed_costs_arr = np.zeros(self.cost_props[0].get_array().shape)
for i in range(len(self.cost_props)):
prop = self.cost_props[i].get_array()
summed_costs_arr += prop * self.cost_weights[i]
self.weight.a = summed_costs_arr
self.time_logs["sum_of_costs"] = round(time.time() - tic, 3)
def remove_vertices(self, dist_surface, delete_padding=0):
"""
Remove edges in a certain corridor (or all) to replace them by
a refined surface
@param dist_surface: a surface where each pixel value corresponds to
the distance of the pixel to the shortest path
@param delete_padding: define padding in which part of the corridor to
delete vertices (cannot delete all because then graph unconnected)
"""
tic = time.time()
self.graph.clear_edges()
self.graph.shrink_to_fit()
self.time_logs["remove_edges"] = round(time.time() - tic, 3)
def get_pareto(self,
vary,
source,
dest,
out_path=None,
compare=[0, 1],
plot=1):
"""
Arguments:
vary: how many weights to explore
e.g 3 --> each cost class can have weight 0, 0.5 or 1
source, dest: as always the source and destination vertex
out_path: where to save the pareto figure(s)
compare: indices of cost classes to compare
Returns:
paths: All found paths
pareto: The costs for each combination of weights
"""
tic = time.time()
# initialize lists
pareto = list()
paths = list()
cost_sum = list()
# get the edge costs
cost_arrs = [cost.get_array() for cost in self.cost_props]
# [self.cost_props[comp].get_array() for comp in compare]
# get vary weights between 0 and 1
var_weights = np.around(np.linspace(0, 1, vary), 2)
# construct weights array
if len(compare) == 2:
weights = [[v, 1 - v] for v in var_weights]
elif len(compare) == 3:
weights = list()
for w0 in var_weights:
for w1 in var_weights[var_weights <= 1 - w0]:
weights.append([w0, w1, 1 - w0 - w1])
else:
raise ValueError("argument compare can only have length 2 or 3")
# w_avail: keep weights of non-compare classes, get leftover amount
w_avail = np.sum(np.asarray(self.cost_weights)[compare])
# compute paths for each combination of weights
for j in range(len(weights)):
# option 2: np.zeros(len(cost_arrs)) + non_compare_weight
w = self.cost_weights.copy()
# replace the ones we want to compare
w[compare] = np.array(weights[j]) * w_avail
# weighted sum of edge costs
self.weight.a = np.sum(
[cost_arrs[i] * w[i] for i in range(len(cost_arrs))], axis=0)
# get shortest path
path, path_costs, _ = self.get_shortest_path(source, dest)
# don't take cost_sum bc this is sum of original weighting
pareto.append(np.sum(path_costs, axis=0)[compare])
paths.append(path)
# take overall sum of costs (unweighted) that this w leads to
cost_sum.append(np.sum(path_costs))
# print best weighting
best_weight = np.argmin(cost_sum)
w = self.cost_weights.copy()
w[compare] = np.array(weights[best_weight]) * w_avail
print("Best weights:", w, "with (unweighted) costs:", np.min(cost_sum))
self.time_logs["pareto"] = round(time.time() - tic, 3)
pareto = np.array(pareto)
classes = [self.cost_classes[comp] for comp in compare]
# Plotting
if plot:
if len(compare) == 2:
plot_pareto_scatter_2d(pareto,
weights,
classes,
cost_sum=cost_sum,
out_path=out_path)
elif len(compare) == 3:
# plot_pareto_3d(pareto, weights, classes)
plot_pareto_scatter_3d(pareto,
weights,
classes,
cost_sum=cost_sum,
out_path=out_path)
return paths, weights, cost_sum
def get_shortest_path(self, source, target):
"""
Compute shortest path from source vertex to target vertex
"""
tic = (time.time())
# #if source and target are given as indices:
if self.graphtool:
vertices_path, _ = shortest_path(self.graph,
source,
target,
weights=self.weight,
negative_weights=True)
else:
try:
vertices_path = nx.dijkstra_path(self.graph, source, target)
except nx.exception.NetworkXNoPath:
return []
self.time_logs["shortest_path"] = round(time.time() - tic, 3)
return vertices_path
def save_graph(self, OUT_PATH):
"""
Save the graph in OUT_PATH
"""
if self.graphtool:
for i, cost_class in enumerate(self.cost_classes):
self.graph.edge_properties[cost_class] = self.cost_props[i]
self.graph.edge_properties["weight"] = self.weight
self.graph.save(OUT_PATH + ".xml.gz")
else:
nx.write_weighted_edgelist(self.graph,
OUT_PATH + '.weighted.edgelist')
def load_graph(self, IN_PATH):
"""
Retrieve graph from IN_PATH
"""
if self.graphtool:
self.g_prev = load_graph(IN_PATH + ".xml.gz")
self.weight_prev = self.g_prev.ep.weight
# weight = G2.ep.weight[G2.edge(66, 69)]
else:
self.g_prev = nx.read_edgelist(IN_PATH + '.weighted.edgelist',
nodetype=int,
data=(('weight', float), ))
# -----------------------------------------------------------------------
# INTERFACE
def single_sp(self, **kwargs):
"""
Function for full processing until shortest path
"""
self.start_inds = kwargs["start_inds"]
self.dest_inds = kwargs["dest_inds"]
self.set_shift(self.start_inds, self.dest_inds, **kwargs)
# self.set_corridor(
# np.ones(self.hard_constraints.shape) * 0.5,
# self.start_inds,
# self.dest_inds,
# factor_or_n_edges=1
# )
if self.verbose:
print("1) Initialize shifts and instance (corridor)")
self.set_edge_costs(**kwargs)
# add vertices
self.add_nodes()
if self.verbose:
print("2) Initialize distances to inf and predecessors")
self.add_edges()
if self.verbose:
print("3) Compute source shortest path tree")
print("number of vertices and edges:", self.n_nodes, self.n_edges)
# weighted sum of all costs
self.sum_costs()
source_v, target_v = self.add_start_and_dest(self.start_inds,
self.dest_inds)
# get actual best path
path, path_costs, cost_sum = self.get_shortest_path(source_v, target_v)
if self.verbose:
print("4) shortest path", cost_sum)
return path, path_costs, cost_sum
def phylomemetic_graph(steps,
communities,
min_size=3,
max_size=50,
parent_limit=2,
workers='auto',
chunksize='auto',
method='fast',
min_backwards_containment=0,
min_forward_containment=0):
'''phylomemetic_graph
Parameters
----------
steps : :obj:`iter` of :obj:`int`
communities : :obj:`iter` of :obj:`iter` of :obj:`int`
min_size : :obj:`int`
max_size : :obj:`int`
parent_limit : :obj:`int`
workers : :obj:`int`
chunksize : :obj:`int`
method : :obj:`str`
min_backwards_containment : :obj:`float`
min_forward_containment : :obj:`float`
Returns
-------
g : :obj:`graph_too.Graph`
group_link_strength : :obj:`graph_tool.EdgePropertyMap`
single_link_strength : :obj:`graph_tool.EdgePropertyMap`
vertex_steps : :obj:`graph_tool.VertexPropertyMap`
element_vertex_map : :obj:`dict`
'''
if workers == 'auto':
workers = cpu_count() - 1
communities_filt = []
communities_lengths = []
element_community_mappings = []
for sequences in communities:
s_filt = list(filter_by_size(sequences, min_size, max_size))
communities_filt.append(s_filt)
communities_lengths.append(len(s_filt))
element_community_mappings.append(reverse_index(s_filt))
community_vertex_maps = []
communities_offsets = []
cumsum_lengths = np.cumsum(communities_lengths)
for length, count in zip(communities_lengths, cumsum_lengths):
start = count - length
end = count
communities_offsets.append((start, end))
community_vertex_maps.append(
{c: v
for c, v in zip(range(length), range(start, end))})
n_communities = np.sum(communities_lengths)
phylomemetic_links = []
for i, (cps, cfs) in enumerate(window(communities_filt, 2)):
n_cf = len(cfs)
logger.info(f'Processing {i+1} of {len(communities)-1} periods')
if chunksize == 'auto':
chunksize_i = int(np.ceil((1 / workers) * n_cf))
else:
chunksize_i = chunksize
with Pool(workers) as pool:
phylomemetic_links.append(
pool.map(
find_links,
zip(
cfs,
range(0, len(cfs)),
repeat(cps, n_cf),
repeat(communities_offsets[i], n_cf),
repeat(element_community_mappings[i], n_cf),
repeat(parent_limit, n_cf),
),
chunksize=chunksize_i,
))
pool.close()
pool.join()
g = Graph(directed=True)
g.add_vertex(n_communities)
group_link_strength = g.new_edge_property('float')
single_link_strength = g.new_edge_property('float')
phylomemetic_links = flatten(flatten(phylomemetic_links))
g.add_edge_list(phylomemetic_links,
eprops=[group_link_strength, single_link_strength])
element_vertex_map = reverse_index_communities(flatten(communities_filt))
vertex_steps = g.new_vertex_property('int')
for (start, end), step in zip(communities_offsets, steps):
vertex_steps.a[start:end] = step
return (g, group_link_strength, single_link_strength, vertex_steps,
element_vertex_map)
class BiblioNetwork():
"Bibliography network displayer"
def __init__(self, filepath):
self.filepath = filepath
self.db = None
self._auth_betw = None
self._auth_betw_computed_from = 0
self.layout_pos = None
self.graph = None
self.author_list = []
@staticmethod
def _split_authors(row):
"Split authors of the row"
auth = row['Authors'].split(", ")
auth = [", ".join(auth[2*i:2*i+2])
for i in range(int(len(auth)/2))]
return auth
def parse(self, nmb_to_import=None, delimiter=","):
"Parse the database csv file"
# import database
self.db = pd.read_csv(self.filepath, delimiter, index_col=False,
nrows=nmb_to_import, encoding="ISO8859",
error_bad_lines=False, warn_bad_lines=True)
self.db.reset_index()
# separate authors
self.db['Authors'] = self.db.apply(self._split_authors, axis=1)
# Replace missing values
self.db['Cited by'].fillna(0, inplace=True)
# Updat author list
self.update_author_list()
def clean(self, min_citations=10):
"Remove some entries"
len_bef = len(self.db)
self.db.drop(self.db[self.db["Cited by"] < min_citations].index,
inplace=True)
len_after = len(self.db)
print(" Removed {} articles, {} remaining".format(len_bef-len_after,
len_after))
self.update_author_list()
self._auth_betw = None
def remove_anterior(self, year):
"Remove some entries"
len_bef = len(self.db)
self.db.drop(self.db[self.db["Year"] <= year].index,
inplace=True)
len_after = len(self.db)
print(" Removed {} articles, {} remaining".format(len_bef-len_after,
len_after))
self.update_author_list()
self._auth_betw = None
def remove_posterior(self, year):
"Remove some entries"
len_bef = len(self.db)
self.db.drop(self.db[self.db["Year"] > year].index,
inplace=True)
len_after = len(self.db)
print(" Removed {} articles, {} remaining".format(len_bef-len_after,
len_after))
self.update_author_list()
self._auth_betw = None
def update_author_list(self):
"Update author list from database"
auths = list(set(np.concatenate(self.db['Authors'].values)))
self.author_list = np.sort(auths)
@property
def author_betweeness(self):
"Compute authors betweness"
# If already computed, just return it
if self._auth_betw is not None and \
self._auth_betw_computed_from == len(self.db):
return self._auth_betw
# else compute it
self._auth_betw_computed_from = len(self.db)
auth_betw = {auth: {}
for auth in self.author_list}
for auths in self.db['Authors']:
# skip if only one author
if len(auths) == 1:
continue
# Loop on authors couples
for i1, auth1 in enumerate(auths):
for auth2 in auths[i1+1::]:
keys = auth_betw.keys()
# create couple if necessary, or increment
if auth2 not in auth_betw[auth1].keys():
auth_betw[auth1][auth2] = 1
else:
auth_betw[auth1][auth2] += 1
if auth1 not in auth_betw[auth2].keys():
auth_betw[auth2][auth1] = 1
else:
auth_betw[auth2][auth1] += 1
self._auth_betw = auth_betw
return self._auth_betw
@author_betweeness.setter
def author_betweeness(self, val):
raise Exception("You cannot change that")
def get_total_citation(self):
""" Return total number of citations for each author"""
nmbcits = {}
for _, art in self.db.iterrows():
auths = art['Authors']
nmbcit = int(art['Cited by'])
for auth in auths:
if auth in nmbcits.keys():
nmbcits[auth] += nmbcit
else:
nmbcits[auth] = nmbcit
return nmbcits
def get_auth_nmb_of_art(self):
""" Return number of article for each author"""
nmbart = {}
for _, art in self.db.iterrows():
auths = art['Authors']
for auth in auths:
if auth in nmbart.keys():
nmbart[auth] += 1
else:
nmbart[auth] = 1
return nmbart
def _get_author_publication(self):
auth2pub = {}
for _, art in self.db.iterrows():
for auth in art['Authors']:
if auth in auth2pub.keys():
auth2pub[auth] += [art.name]
else:
auth2pub[auth] = [art.name]
return auth2pub
def write_author_list(self, filepath):
with open(filepath, "w") as f:
data = ['{}: {}\n'.format(i, auth)
for i, auth in enumerate(self.author_list)]
f.writelines(data)
def make_article_graph(self, layout="arf"):
"""Make an article graph"""
self.graph = Graph(directed=False)
# add vertex
self.graph.add_vertex(len(self.db))
# add properties
cb = self.graph.new_vertex_property("int", self.db['Cited by'].values)
self.graph.vertex_properties['nmb_citation'] = cb
# Add links
auths = list(self.author_betweeness.keys())
auth2ind = {auths[i]: i
for i in range(len(auths))}
auth2pub = self._get_author_publication()
for _, pubs in auth2pub.items():
if len(pubs) < 2:
continue
combis = itertools.combinations(pubs, 2)
self.graph.add_edge_list(list(combis))
# layout
if layout == "arf":
self.layout_pos = arf_layout(self.graph)
elif layout == "sfpd":
self.layout_pos = sfdp_layout(self.graph)
elif layout == "fr":
self.layout_pos = fruchterman_reingold_layout(self.graph)
elif layout == "radial":
self.layout_pos = radial_tree_layout(self.graph,
auth2ind['Logan, B.E.'])
else:
raise ValueError()
def make_author_graph(self, layout="arf"):
"""Make an author graph"""
self.graph = Graph(directed=False)
# add vertex
auths = self.author_list
self.graph.add_vertex(len(auths))
# add links
auth2ind = {auths[i]: i
for i in range(len(auths))}
abet = []
authbet = copy.deepcopy(self.author_betweeness)
for auth in auths:
for col, weight in authbet[auth].items():
if col == auth:
continue
self.graph.add_edge(auth2ind[auth], auth2ind[col])
del authbet[col][auth] # ensure that edges are not doubled
abet.append(weight)
# add properties
cb = self.graph.new_edge_property("int", abet)
self.graph.edge_properties['weight'] = cb
# layout
if layout == "arf":
self.layout_pos = arf_layout(self.graph,
weight=self.graph.ep.weight,
pos=self.layout_pos,
max_iter=10000)
elif layout == "sfpd":
self.layout_pos = sfdp_layout(self.graph,
eweight=self.graph.ep.weight,
pos=self.layout_pos)
elif layout == "fr":
self.layout_pos = fruchterman_reingold_layout(self.graph,
weight=self.graph.ep.weight,
circular=True,
pos=self.layout_pos)
elif layout == "radial":
nc = self.get_total_citation()
main_auth_ind = np.argmax(list(nc.values()))
main_auth = list(nc.keys())[main_auth_ind]
self.layout_pos = radial_tree_layout(self.graph,
auth2ind[main_auth])
elif layout == "planar":
self.layout_pos = planar_layout(self.graph)
else:
raise ValueError()
def display_article_graph(self, out="graph.pdf", min_size=1,
max_size=10, indice=False):
"""Display an article graph
One point per article.
Size and color corespond to the number of citation.
"""
cb = np.log(np.array(self.graph.vp.nmb_citation.a)+2)
ms = cb/max(cb)*(max_size - min_size) + min_size
ms = self.graph.new_vertex_property('float', ms)
graph_draw(self.graph, pos=self.layout_pos, output=out,
vertex_size=ms,
vertex_fill_color=self.graph.vp.nmb_citation,
vcmap=plt.cm.viridis)
def display_author_graph(self, out="graph.pdf", min_size=1, max_size=10,
indice=False):
"""Display an author graph """
auths = self.author_list
nc = self.get_total_citation()
nc = [int(nc[auth]) for auth in auths]
na = self.get_auth_nmb_of_art()
na = [int(na[auth]) for auth in auths]
# normalize citation number
nc = np.array(nc, dtype=float)
nc /= np.max(nc)
nc *= (max_size - min_size)
nc += min_size
# normalize edge width
weight = np.array(self.graph.ep.weight.a, dtype=float)
weight /= np.max(weight)
weight *= (1 - 0.1)
weight += 0.1
# Get vertex display order
vorder = np.argsort(nc)
# Get index
if indice:
text = range(len(vorder))
textg = self.graph.new_vertex_property('string', text)
else:
textg = None
# plot
ncg = self.graph.new_vertex_property('float', nc)
nag = self.graph.new_vertex_property('int', na)
vorderg = self.graph.new_vertex_property('int', vorder)
weightg = self.graph.new_edge_property('float', weight)
self.graph.vp['nmb_citation'] = ncg
graph_draw(self.graph, pos=self.layout_pos, output=out,
vertex_fill_color=nag, vertex_size=ncg,
edge_pen_width=weightg, vertex_text=textg,
vorder=vorderg,
vertex_text_position=0,
vcmap=plt.cm.PuBu)
def test_find_minimum_branching(g, weights):
edges = find_minimum_branching(g, weights=weights)
tree = Graph(directed=True)
tree.add_edge_list(edges)
assert is_arborescence(tree)
