ppt = time.time()
f = open(DATADIR + DATASET + '\\' + FILE_NAME, 'w')
pt = time.time()
net = Graph(DATADIR + DATASET + '\\links.txt', typ='dir', order=ORDER, withdiag=WITHDIAG)
print('READ TIME: %.2f' % (time.time() - pt))
f.write('%d %d %d\n' % (net.nVertices, net.nEdges, DIMENSION))
pt = time.time()
grouping_model = Louvain(net, rand=RANDOM_GROUPING)
groups = grouping_model.execute(merge=MERGE)
print('GROUP TIME: %.2f' % (time.time() - pt))
inv_index_original = groups2inv_index(groups, net.nVertices)
pt = time.time()
k_set = sample(net, k=K_SIZE, method=SAMPLE_METHOD)
print('SAMPLE TIME: %.2f' % (time.time() - pt))
inv_index = groups2inv_index(groups, net.nVertices, k_set)
pure_override_nodes(groups, inv_index)
groups = [k_set] + groups
pt = time.time()
model = Optimizer(net, groups, dim=DIMENSION, lam=LAMBDA, eta=ETA, max_iter=MAX_ITER,
sample_strategy=SAMPLE_METHOD, verbose=True)
print('INITIAL OPTIMIZER TIME (SVD): %.2f' % (time.time() - pt))
with Pool(processes=4) as pool:
random.shuffle(new_interest_set)
groups = []
group = []
for node in new_interest_set:
group.append(node)
if len(group) >= MERGE[1]:
groups.append(group)
group = []
pt = time.time()
k_set = sample(net, k=K_SIZE, method=SAMPLE_METHOD)
sample_time = time.time() - pt
print('SAMPLE TIME: %.2f' % sample_time)
inv_index = groups2inv_index(groups, net.nVertices, k_set)
pure_override_nodes(groups, inv_index)
groups = [k_set] + groups
pt = time.time()
optimizer = Optimizer(net,
groups,
dim=DIMENSION,
lam=LAMBDA,
eta=ETA,
max_iter=MAX_ITER,
sample_strategy=SAMPLE_METHOD,
verbose=VERBOSE)
svd_time = time.time() - pt
print('INITIAL OPTIMIZER TIME (SVD): %.2f' % svd_time)
def process(args):
pt = time.time()
WITHDIAG = True if args.order == 1 else False
net = Graph(args.input, typ='dir', order=args.order,
withdiag=WITHDIAG, verbose=(args.v > 1))
read_time = time.time() - pt
if args.v:
print('READ TIME:\t%.2f' % read_time)
pt = time.time()
grouping_model = Louvain(net, rand=True, verbose=(args.v > 1))
groups = grouping_model.execute(merge=(args.merge0, args.merge1))
group_time = time.time() - pt
if args.v:
print('GROUP TIME:\t%.2f' % group_time)
SAMPLE_METHOD = 'set_cover_undir'
if args.sample == 1:
SAMPLE_METHOD = 'deg_deter'
elif args.sample == 2:
SAMPLE_METHOD = 'deg_prob'
elif args.sample == 3:
SAMPLE_METHOD = 'deg^2_prob'
elif args.sample == 4:
SAMPLE_METHOD = 'uniform'
pt = time.time()
k_set = sample(net, k=args.ksize, method=SAMPLE_METHOD)
sample_time = time.time() - pt
if args.v:
print('SAMPLE TIME:\t%.2f' % sample_time)
inv_index = groups2inv_index(groups, net.nVertices, k_set)
pure_override_nodes(groups, inv_index)
groups = [k_set] + groups
pt = time.time()
optimizer = Optimizer(net, groups, dim=args.dim, lam=args.lam, eta=args.eta,
max_iter=args.iter, sample_strategy=SAMPLE_METHOD,
verbose=(args.v > 1))
svd_time = time.time() - pt
if args.v:
print('INITIAL OPTIMIZER TIME (SVD):\t%.2f' % svd_time)
pt = time.time()
branches = []
for t in range(len(groups)):
branches.append(BranchOptimizer(optimizer, t, verbose=(args.v > 1)))
prep_time = time.time() - pt
if args.v:
print('PROCESS PREPARATION TIME:\t%.2f' % prep_time)
pt = time.time()
with Pool(processes=args.workers) as pool:
grouped_embeddings = pool.map(WrapTrain, branches)
embed_time = time.time() - pt
print('OPTIMIZING TIME:\t%.2f' % embed_time)
if args.v:
total_time = read_time + group_time + sample_time + \
svd_time + prep_time + embed_time
print('TOTAL TIME:\t%.2f' % total_time)
if args.output is not None:
filename = args.output + '_' + '_'.join(
['ksize=%d' % args.ksize,
'dim=%d' % args.dim,
'lam=%.1f' % args.lam,
'eta=%.1f' % args.eta,
'iter=%d' % args.iter
])
f = open(filename, 'w')
f.write('|V|=%d; |E|=%d; dim=%d\n' % (net.nVertices, net.nEdges, args.dim))
for i, group in enumerate(groups):
embeddings = grouped_embeddings[i][1]
for j, newVid in enumerate(group):
vid = net.newVid2vid_mapping[newVid]
f.write('%d ' % vid)
vec = np.array(embeddings[:, j])
vec_str = ' '.join([str(t) for t in vec[:, 0]])
f.write(vec_str)
f.write('\n')
f.close()
pt = time.time()
net = Graph(DATADIR + DATASET + '\\links.txt',
typ=TYPE,
order=ORDER,
withdiag=WITHDIAG,
verbose=VERBOSE)
print('READ TIME: %.2f' % (time.time() - pt))
pt = time.time()
grouping_model = Louvain(net, verbose=VERBOSE)
groups = grouping_model.execute(merge=MERGE)
print('GROUP TIME: %.2f' % (time.time() - pt))
group_sizes = [len(t) for t in groups]
vid2originalGroup_index = groups2inv_index(groups, net.nVertices)
vid2originalGroupSize_index = [
group_sizes[t - 1] for t in vid2originalGroup_index
]
pt = time.time()
k_set = sample(net,
k=K_SIZE,
method=SAMPLE_METHOD,
vertex_group_sizes=vid2originalGroupSize_index)
print('SAMPLE TIME: %.2f' % (time.time() - pt))
inv_index = groups2inv_index(groups, net.nVertices, k_set)
pure_override_nodes(groups, inv_index)
groups = [k_set] + groups
