def test_search():
a = TimeCollection([((1, False), 3), ((4, True), 4)], discrete=False)
for i, v1, v2 in zip(range(6), 4 * [0] + 2 * [1], 5 * [0] + [1]):
assert_equal(a.search_time(i), v1)
assert_equal(a.search_time((i, False)), v2)
a = TimeCollection([(1, 3), (4, 4)], discrete=True)
for i, v1 in zip(range(6), 4 * [0] + 2 * [1]):
assert_equal(a.search_time(i), v1)
a = TimeCollection([(1, 3), (4, 4)], instantaneous=True)
for i, v1 in zip(range(6), [None, 0, None, None, 1, None]):
assert_equal(a.search_time(i), v1)
def nodes_at(self, t=None):
if bool(self):
if t is None:
def generate(iter_):
prev = None
for u, ts in iter_:
if prev is None:
active_set, prev = {u}, ts
elif ts != prev:
yield (prev, NodeSetS(set(active_set)))
active_set, prev = {u}, ts
else:
active_set.add(u)
if len(active_set):
yield (prev, NodeSetS(set(active_set)))
# Iterate in ascending time and yield the NodeSet at each time-instant in a generator fashion
return TimeGenerator(generate(self.sort_df('ts').itertuples()),
instantaneous=True,
discrete=self.discrete)
elif isinstance(t, Real):
# Count how many times its time-step occurs, for each key and return a sorted list
return NodeSetS(self.df.df_at(t).u)
else:
raise ValueError(
'Input can either be a real number or an ascending interval of two real numbers'
)
else:
if t is None:
return TimeCollection(iter(),
instantaneous=True,
discrete=self.discrete)
else:
return NodeSetS()
def link_contribution_at(self, t=None):
"""Calculate the contribution of a link inside the stream_graph.
Parameters
----------
t: time or None
Returns
-------
node_coverage : Real or TimeCollection
Returns :math:`l_{t}=\\frac{|E_{t}|}{|V*(V-1)|}`.
If None returns the time coverage for each node at each time-event.
"""
denom = float(self.nodeset_.size)
denom = denom * (denom - 1)
if t is None:
if denom > .0:
def fun(t, v):
return v / denom
return self.temporal_linkset_.m_at().map(fun)
else:
return TimeCollection(
instants=self.temporal_linkset_.instantaneous)
else:
if denom > .0:
return self.temporal_linkset_.m_at(t) / denom
else:
return .0
def _m_at_unweighted(self, t):
if t is None:
from stream_graph.collections import TimeCollection
time_links = self.links_at()
return TimeCollection([(time, l.size) for time, l in time_links],
time_links.instants)
else:
return self.links_at(t).size
def _m_at_weighted(self, t):
if t is None:
return TimeCollection(self._build_time_generator(
Counter, sum_counter_),
instantaneous=False,
discrete=self.discrete)
else:
return self.df.count_at(t, weights=True)
def _m_at_weighted(self, t):
if t is None:
ct = Counter()
for ts, w in self.df[['ts', 'w']].itertuples(weights=True):
ct[ts] += w
# Add up weigths for each time-stamp and sort on time.
return TimeCollection(sorted(iteritems(ct)), instantaneous=True, discrete=self.discrete)
else:
return self.links_at(t).weighted_size
def _m_at_unweighted(self, t):
if t is None:
return TimeCollection(self._build_time_generator(set,
len_set_,
weighted=False),
instantaneous=False,
discrete=self.discrete)
else:
return self.df.count_at(t)
def _degree_at_weighted(self, u, t, direction):
if u is None:
if t is None:
out = dict()
for u, val in self._build_time_generator(
Counter,
sum_counter_n,
direction=direction,
get_key=get_key_first):
d = out.get(u, None)
if d is None:
out[u] = TimeCollection([val],
discrete=self.discrete,
instantaneous=False)
else:
d.append(val)
return NodeCollection(out)
else:
return LinkSetDF(self.df.df_at(t)[['u', 'v', 'w']],
weighted=True,
merge_function=self.df_.merge_function,
no_duplicates=False).degree(
u=None, direction=direction, weights=True)
else:
if direction == 'out':
df = self.df[self.df.u == u].drop(
columns=['u'], merge=False).rename(columns={'v': 'u'})
elif direction == 'in':
df = self.df[self.df.v == u].drop(columns=['v'], merge=False)
elif direction == 'both':
df = self.df[self.df.u == u].drop(
columns=['u'], merge=False).rename(columns={'v': 'u'})
df = df.append(self.df[self.df.v == u].drop(columns=['v'],
merge=True),
ignore_index=True)
else:
raise UnrecognizedDirection()
if t is None:
return TimeCollection(self._build_time_generator(
Counter, sum_counter_, direction=direction, df=df),
discrete=self.discrete,
instantaneous=False)
else:
return df.w[df.index_at(t)].sum()
def neighbors_at(self, u=None, t=None, direction='out'):
if not bool(self):
if u is None:
return NodeCollection()
if t is None:
return TimeCollection()
return NodeSetS()
mf = (self.df_.merge_function if self.weighted else None)
if u is None:
if t is None:
out = dict()
for u, val in self._build_time_generator(
set,
set_unweighted_n_sparse,
weighted=False,
direction=direction,
get_key=get_key_first,
sparse=True):
d = out.get(u, None)
if d is None:
out[u] = TimeSparseCollection([val],
discrete=self.discrete,
caster=NodeSetS)
else:
d.append(val)
return NodeCollection(out)
else:
return LinkSetDF(self.df.df_at(t)[['u', 'v'] + self._wc],
merge_function=mf,
no_duplicates=False).neighbors_of(
u=None, direction=direction)
else:
if direction == 'out':
df = self.df[self.df.u == u].drop(
columns=['u'] + self._wc,
merge=False).rename(columns={'v': 'u'})
elif direction == 'in':
df = self.df[self.df.v == u].drop(columns=['v'] + self._wc,
merge=False)
elif direction == 'both':
df = self.df[self.df.u == u].drop(
columns=['u'] + self._wc).rename(columns={'v': 'u'})
df = df.append(self.df[self.df.v == u].drop(columns=['v'] +
self._wc,
merge=True),
ignore_index=True)
else:
raise UnrecognizedDirection()
if t is None:
return TemporalNodeSetDF(df).nodes_at(t=None)
else:
return NodeSetS(df[df.index_at(t)].u.values.flat)
def n_at(self, t=None):
if bool(self):
if t is None:
return self._build_time_generator(set, len_set_nodes,
TimeCollection)
else:
return self.df.count_at(t)
else:
if t is None:
return TimeCollection([], False)
else:
return 0
def n_at(self, t=None):
if bool(self):
if t is None:
# Count how many times its time-step occurs, for each key and return a sorted list
return TimeCollection(sorted(
list(iteritems(Counter(t for t in self.df.ts)))),
instantaneous=True,
discrete=self.discrete)
elif isinstance(t, Real):
# Count for only one time-stamp
return len(set(self.df.df_at(t).u))
else:
raise ValueError(
'Input can either be a real number or an ascending interval of two real numbers'
)
else:
if t is None:
return TimeCollection(iter(),
instantaneous=True,
discrete=self.discrete)
else:
return NodeSetS()
def n_at(self, t=None):
if t is None:
if bool(self):
return constant_time_generator(self.timeset, self.n, 0,
self.discrete,
self.instantaneous)
else:
return TimeCollection()
else:
if bool(self) and (t in self.timeset_):
return self.n
else:
return 0
def n_at(self, t=None):
"""Returns number of nodes of the TemporalNodeSet at a certain time.
Parameters
----------
t : Real
Returns
-------
n : Int or TimeCollection(Int)
Returns the number of active nodes at time :code:`t`.
If None returns an iterator of tuples containing a timestamp and an Int.
"""
if t is None:
from stream_graph.collections import TimeCollection
time_nodes = self.nodes_at()
return TimeCollection([(ts, ns.size) for (ts, ns) in time_nodes],
time_nodes.instants)
else:
return self.nodes_at(t).size
def _degree_at_unweighted(self, u, t, direction):
if u is None:
if t is None:
out = dict()
for u, val in self._build_time_generator(
set, len_set_n, direction=direction,
get_key=get_key_first):
d = out.get(u, None)
if d is None:
out[u] = TimeCollection([val],
discrete=self.discrete,
instantaneous=False)
else:
d.append(val)
return NodeCollection(out)
else:
return LinkSetDF(self.df.df_at(t)[['u', 'v'] + self._wc],
no_duplicates=False).degree(
u=None, direction=direction)
else:
df = (self.df.drop(columns='w', merge=False)
if self.weighted else self.df)
if direction == 'out':
df = df[df.u == u].drop(
columns=['u'],
merge=self.weighted).rename(columns={'v': 'u'})
elif direction == 'in':
df = df[df.v == u].drop(columns=['v'], merge=self.weighted)
elif direction == 'both':
dfa = df[df.u == u].drop(
columns=['u'], merge=False).rename(columns={'v': 'u'})
df = dfa.append(df[df.v == u].drop(columns=['v'], merge=False),
ignore_index=True,
merge=True)
else:
raise UnrecognizedDirection()
if t is None:
return TemporalNodeSetDF(df).n_at(t=None)
else:
return len(set(df.df_at(t).u.values.flat))
def m_at(self, t=None, weights=False):
"""Returns the number of links appearing at a certain time.
Parameters
----------
t : Int
weights : bool, default=False
Returns
-------
m : Int
Returns the numer of links at a certain time.
"""
if not bool(self):
if t is None:
return TimeCollection(discrete=self.discrete_)
else:
return .0
elif weights and self.weighted:
return self._m_at_weighted(t)
else:
return self._m_at_unweighted(t)
def _degree_at_unweighted(self, u=None, t=None, direction='out'):
if not bool(self):
if u is None:
return NodeCollection()
if t is None:
return TimeCollection(discrete=self.discrete, instantaneous=True)
return 0
if u is None:
if t is None:
out = dict()
if direction == 'out':
def add(d, u, v):
d[u].add(v)
elif direction == 'in':
def add(d, u, v):
d[v].add(u)
elif direction == 'both':
def add(d, u, v):
d[u].add(v)
d[v].add(u)
else:
raise UnrecognizedDirection()
prev = None
for u, v, ts in self.sort_df('ts').itertuples():
# Collect neighbors at each time-stamp
if prev is None:
cache = defaultdict(set)
prev = ts
elif ts != prev:
for z, s in iteritems(cache):
if z in out:
# and calculate their size
out[z].it.append((prev, len(s)))
else:
# in a TimeCollection of ascending time for each node
out[z] = TimeCollection([(prev, len(s))], discrete=self.discrete, instantaneous=True)
cache = defaultdict(set)
prev = ts
add(cache, u, v)
# Add the remaining, from the cache
for u, s in iteritems(cache):
if u in out:
out[u].it.append((prev, len(s)))
else:
out[u] = TimeCollection([(prev, len(s))], discrete=self.discrete, instantaneous=True)
return NodeCollection(out)
else:
return LinkSetDF(self.df.df_at(t).drop(columns=['ts']), weighted=self.weighted).degree(u=None, direction=direction)
else:
if direction == 'out':
df = self.df[self.df.u == u].drop(columns=['u'], merge=False).rename(columns={'v': 'u'})
elif direction == 'in':
df = self.df[self.df.v == u].drop(columns=['v'], merge=False)
elif direction == 'both':
df = self.df[self.df.u == u].drop(columns=['u'], merge=False).rename(columns={'v': 'u'})
df = df.append(self.df[self.df.v == u].drop(columns=['v'], merge=False), ignore_index=True, merge=True)
else:
raise UnrecognizedDirection()
if t is None:
dt = defaultdict(set)
# Collect all nodes for each time-stamp
for u, ts in df.itertuples():
dt[ts].add(u)
return TimeCollection(sorted(list((ts, len(us)) for ts, us in iteritems(dt))), discrete=self.discrete, instantaneous=True)
else:
return len(set(df.df_at(t).u.values.flat))
def neighbors_at(self, u=None, t=None, direction='out'):
if not bool(self):
if u is None:
return NodeCollection()
if t is None:
return TimeCollection(discrete=self.discrete, instantaneous=True)
return NodeSetS()
if u is None:
if t is None:
out = dict()
if direction == 'out':
def add(d, u, v):
d[u].add(v)
elif direction == 'in':
def add(d, u, v):
d[v].add(u)
elif direction == 'both':
def add(d, u, v):
d[u].add(v)
d[v].add(u)
else:
raise UnrecognizedDirection()
prev = None
for u, v, ts in self.sort_df('ts').itertuples():
# Iterate in ascending time and at each instant collect for each node the set of its neighbors
if prev is None:
prev, cache = ts, defaultdict(set)
elif ts != prev:
# For each node
for z, s in iteritems(cache):
if z in out:
# Or append it if it exists
out[z].it.append((prev, NodeSetS(s)))
else:
# Initialize a time-collection with the time-stamp and the node-set
out[z] = TimeCollection([(prev, NodeSetS(s))], instantaneous=True, discrete=self.discrete)
prev, cache = ts, defaultdict(set)
add(cache, u, v)
# Add also remaining elements.
for u, s in iteritems(cache):
if u in out:
out[u].it.append((prev, NodeSetS(s)))
else:
out[u] = TimeCollection([(prev, NodeSetS(s))], instantaneous=True, discrete=self.discrete)
return NodeCollection(out)
else:
return LinkSetDF(self.df.df_at(t).drop(columns=['ts']), weighted=self.weighted).neighbors_of(u=None, direction=direction)
else:
di = True
if direction == 'out':
df = self.df[self.df.u == u].drop(columns=['u']).rename(columns={'v': 'u'})
elif direction == 'in':
df = self.df[self.df.v == u].drop(columns=['v'])
elif direction == 'both':
df = self.df[self.df.u == u].drop(columns=['u']).rename(columns={'v': 'u'})
df = df.append(self.df[self.df.v == u].drop(columns=['v']), ignore_index=True)
di = False
else:
raise UnrecognizedDirection()
if t is None:
return ITemporalNodeSetDF(df[['u', 'ts']], no_duplicates=di, discrete=self.discrete).nodes_at(t=None)
else:
return NodeSetS(df.df_at(t).u.values.flat)
def _m_at_unweighted(self, t):
if t is None:
# Count how many times each time-stamp occurs and sort on time.
return TimeCollection(sorted(list(iteritems(Counter(iter(self.df.ts))))), instantaneous=True, discrete=self.discrete)
else:
return self.links_at(t).size
def test_merge():
# Continuous
tc_a = TimeCollection([((1, True), 4), ((2, True), 3), ((4, False), 5), ((5, False), 0), ((6, False), 1), ((7, True), 0)])
tc_b = TimeCollection([((1.5, False), 1), ((2, False), 2), ((3, False), 1), ((5.5, True), 0), ((8, True), 1)])
assert_equal(list(tc_a.merge(tc_b, add, missing_value=0)), list(tc_b.merge(tc_a, add, missing_value=0)))
assert_equal(list(tc_a.merge(tc_b, add, missing_value=0)), [((1, True), 4), ((1.5, False), 5), ((2, False), 6), ((2, True), 5), ((3, False), 4), ((4, False), 6), ((5, False), 1), ((5.5, True), 0), ((6, False), 1), ((7, True), 0), ((8, True), 1)])
assert not tc_a.merge(tc_b, add, missing_value=0).instants
# future fix
# tc_a = TimeCollection([((1, True), 4), ((2, True), 3), ((4, False), 5), ((5, False), 0), ((6, False), 1), ((7, True), 0)])
# tc_b = TimeCollection([(1.5, 1), (2, 2), (3, 1), (5.5, 0), (8, 1)], instantaneous=True)
# print(tc_a.merge(tc_b, add, missing_value=0))
# assert list(tc_a.merge(tc_b, add, missing_value=0)) == list(tc_b.merge(tc_a, add, missing_value=0))
# assert list(tc_a.merge(tc_b, add, missing_value=0)) == [(1.5, 5), (2, 5), (3, 4), (5.5, 0), (8, 1)]
# assert not tc_a.merge(tc_b, add, missing_value=0).instants
# tc_a = TimeCollection([(1, 4), (2, 3), (4, 5), ((5, False), 0), ((6, False), 1), ((7, True), 0)], instantaneous=True)
# tc_b = TimeCollection([(1.5, 1), (2, 2), (3, 1), (5.5, 0), (8, 1)], instantaneous=True)
# assert list(tc_a.merge(tc_b, add, missing_value=0)) == list(tc_b.merge(tc_a, add, missing_value=0))
# assert list(tc_a.merge(tc_b, add, missing_value=0)) == [(1, 4), (1.5, 1), (2, 5), (3, 1), (4, 5), (5, 0), (5.5, 0), (6, 1), (7, 0), (8, 1)]
# assert tc_a.merge(tc_b, add, missing_value=0).instants
# Discrete
tc_a = TimeCollection([(2, 4), (4, 3), (8, 5), (10, 0), (12, 1), (14, 0)], discrete=True)
tc_b = TimeCollection([(3, 1), (4, 2), (6, 1), (11, 0), (16, 1)], discrete=True)
assert_equal(list(tc_a.merge(tc_b, add, missing_value=0)), list(tc_b.merge(tc_a, add, missing_value=0)))
assert_equal(list(tc_a.merge(tc_b, add, missing_value=0)), [(2, 4), (3, 5), (6, 4), (8, 6), (10, 1), (11, 0), (12, 1), (14, 0), (16, 1)])
assert not tc_a.merge(tc_b, add, missing_value=0).instants
# tc_a = TimeCollection([(2, 4), (3, 3), (4, 5), (10, 0), (12, 1), (14, 0)], instantaneous=True)
# tc_b = TimeCollection([(3, 1), (4, 2), (6, 1), (11, 0), (16, 1)], discrete=True, instantaneous=True)
# assert list(tc_a.merge(tc_b, add, missing_value=0)) == list(tc_b.merge(tc_a, add, missing_value=0))
# assert list(tc_a.merge(tc_b, add, missing_value=0)) == [(1.5, 5), (2, 5), (3, 4), (5.5, 0), (8, 1)]
# assert not tc_a.merge(tc_b, add, missing_value=0).instants
tc_a = TimeCollection([(2, 4), (4, 3), (8, 5), (10, 0), (12, 1), (14, 0)], discrete=True, instantaneous=True)
tc_b = TimeCollection([(3, 1), (4, 2), (6, 1), (11, 0), (16, 1)], discrete=True, instantaneous=True)
assert_equal(list(tc_a.merge(tc_b, add, missing_value=0)), list(tc_b.merge(tc_a, add, missing_value=0)))
assert_equal(list(tc_a.merge(tc_b, add, missing_value=0)), [(2, 4), (3, 1), (4, 5), (6, 1), (8, 5), (10, 0), (11, 0), (12, 1), (14, 0), (16, 1)])
assert tc_a.merge(tc_b, add, missing_value=0).instants
def ego(e, ne, l, both, detailed, discrete):
# print >> sys.stderr, "Running node : " + str(e)
u, v, t, index, times = 0, 0, 0, 0, list()
ce, info = dict(), dict()
for i in ne:
info[(e, i)] = -1
for x in ne - {i}:
info[(i, x)] = -1
info[(i, e)] = -1
index = 0
time = l[index][2] # starting time.
ne_x, lines, paths = ne | {e}, dict(), dict()
if both:
def add_lines(u, v, t):
lines[(u, v)] = t
lines[(v, u)] = t
else:
def add_lines(u, v, t):
lines[(u, v)] = t
if detailed:
def take(times, prev, val):
times.append((time, val))
else:
if discrete:
def take(times, prev, val):
if prev[0] is None or prev[0] != val:
times.append((time, val))
prev[0] = val
else:
def take(times, prev, val):
if prev[0] is None or prev[0] != val:
times.append(((time, True), val))
prev[0] = val
prev = [None]
while (index < len(l) - 1):
# get all links of time stamp
while (index < len(l) - 1):
u, v, t = l[index]
add_lines(u, v, t)
index += 1
if (t != time):
break
for u in ne:
for v in ne - {u}:
# print u,v
ce[(u, v)] = 0.0
Q = set()
if (u, v) not in lines:
# print u,v
news = info[(u, v)]
for x in ne_x - {u, v}:
ux = info[(u, x)]
if (x, v) in lines:
xv = lines[(x, v)]
else:
xv = info[(x, v)]
if ux != -1 and xv != -1 and ux < xv:
if ux == news:
Q.add(x)
if ux > news:
Q = {x}
news = ux
if (u, v) in paths:
old_paths = paths[(u, v)]
if old_paths[0] == news:
paths[(u, v)] = (news, paths[(u, v)][1] | Q)
elif old_paths[0] < news:
paths[(u, v)] = (news, Q)
else:
paths[(u, v)] = (news, Q)
if e in paths[(u, v)][1]:
ce[(u, v)] = 1.0 / len(paths[(u, v)][1])
else:
paths[(u, v)] = (t, {u})
val = sum(ce.values())
take(times, prev, val)
for k in lines:
info[k] = lines[k]
time, lines = l[index][2], {}
return TimeCollection(times, instantaneous=detailed, discrete=discrete)
def _degree_at_weighted(self, u, t, direction):
if u is None:
if t is None:
out = dict()
if direction == 'out':
def add(d, u, v, w):
d[u] += w
elif direction == 'in':
def add(d, u, v, w):
d[v] += w
elif direction == 'both':
def add(d, u, v, w):
d[u] += w
d[v] += w
else:
raise UnrecognizedDirection()
prev = None
for u, v, ts, w in self.sort_df('ts').itertuples(weights=True):
# Iterate in ascending time
if prev is None:
# For each node add all the weights for all its neighbors
cache = Counter()
prev = ts
elif ts != prev:
for z, weight in iteritems(cache):
if z in out:
# Append in ascending time inside the TimeCollection
out[z].it.append((prev, weight))
else:
# Initialize inside the TimeCollection
out[z] = TimeCollection([(prev, weight)], discrete=self.discrete, instantaneous=True)
cache = Counter()
prev = ts
add(cache, u, v, w)
# Remove the remaining.
for u, weight in iteritems(cache):
if u in out:
out[u].it.append((prev, weight))
else:
out[u] = TimeCollection([(prev, weight)], discrete=self.discrete, instantaneous=True)
return NodeCollection(out)
else:
return LinkSetDF(self.df.df_at(t).drop(columns=['ts']), weighted=self.weighted, merge_function=self.df_.merge_function).degree(u=None, direction=direction, weights=True)
else:
if direction == 'out':
df = self.df[self.df.u == u].drop(columns=['u'], merge=False).rename(columns={'v': 'u'})
elif direction == 'in':
df = self.df[self.df.v == u].drop(columns=['v'], merge=False)
elif direction == 'both':
df = self.df[self.df.u == u].drop(columns=['u'], merge=False).rename(columns={'v': 'u'})
df = df.append(self.df[self.df.v == u].drop(columns=['v'], merge=False), ignore_index=True, merge=True)
else:
raise UnrecognizedDirection()
if t is None:
dt = Counter()
for u, ts, w in df.itertuples(weights=True):
# collect all the weights for its time-stamp.
dt[ts] += w
return TimeCollection(sorted(list(iteritems(dt))), discrete=self.discrete, instantaneous=True)
else:
return df.df_at(t).w.sum()
