def main(data_name, r=1):
with open(data_name + '.pkl', 'rb') as f:
data = pickle.load(f)
insts = data.keys()
seeds = list(filter(lambda x: '#' not in x, insts))
for i in seeds:
inst = TSP(data[i])
dists = np.array(inst.dist_mat).flatten()
# normalise to r == 1
dists = dists / dists.max() * 2.0 * r
_, shp, _, scl = exponweib.fit(dists, f0=1)
print(i, np.abs(-0.509 * shp + 0.707))
# plt.clf()
# plt.title(i)
# _ = plt.hist(dists, density=True)
# x = np.linspace(dists.min(), dists.max(), 1000)
# plt.plot(x, weibull(x, shp, scl))
# plt.show()
for i in range(5):
inst = TSP(int(data_name.split('_')[-1]))
dists = np.array(inst.dist_mat).flatten()
dists = dists / dists.max() * 2.0 * r
_, shp, _, scl = exponweib.fit(dists, f0=1)
print('rand' + str(i), np.abs(-0.509 * shp + 0.707))
def cat_exp_main(data_dir):
m = 3
metric = Map_Dist()
fns_ = os.listdir(data_dir)
fns = list(filter(lambda s: s.endswith('.pkl'), fns_))
for fn in fns:
ofn = fn[:-3] + 'csv'
if ofn in fns_:
continue
with open(os.path.join(data_dir, fn), 'rb') as f:
data = pickle.load(f)
insts = data.keys()
s = min(insts, key=len)
print(s, end='')
seed = TSP(data[s])
with open(os.path.join(data_dir, ofn), 'w') as f:
print('Inst_a,Inst_b,Distance,Baseline', file=f)
for i in insts:
print('*', end='')
inst = TSP(data[i])
if s == i:
t = s + '_r0_#0'
else:
t = i
for _ in range(m):
res = (s, t, str(metric.dist(seed, inst)['dist']),
str(abs(seed.hardness_est() - inst.hardness_est())))
with open(os.path.join(data_dir, ofn), 'a') as f:
print(','.join(res), file=f)
print()
def init_exp_main(data_dir):
metrics = [
Map_Dist(),
Geo_Dist(),
Geo_Dist(embedding='deepwalk'),
Geo_Dist(embedding='node2vec'),
# Geo_Dist(embedding='line'),
# Geo_Dist(embedding='sdne'),
Geo_Dist(embedding='struc2vec'),
Abstract_Dist(),
DMST_Heuristic(),
FI_Heuristic(),
# Gr_Heuristic(),
NI_Heuristic(),
NN_Heuristic()
]
solver = GA_Solver()
ofn = 'init_0.txt'
with open(ofn, 'w') as f:
print('inst_a inst_b best', file=f, end=' ')
for l in [m.lbl for m in metrics]:
print(l, file=f, end=' ')
print(file=f)
fns_ = os.listdir(data_dir)
fns = list(filter(lambda s: s.endswith('.pkl'), fns_))
for fn in fns:
with open(os.path.join(data_dir, fn), 'rb') as f:
data = pickle.load(f)
keys = list(data.keys())
i = np.argmin([len(k) for k in keys])
inst_a = TSP(data[keys[i]])
print(keys[i])
for k in keys:
inst_b = TSP(data[k])
res = []
for m in metrics:
if m.lbl.startswith('heuristic'):
res.append(
solver.solve(inst_b,
seed=m.solve(inst_b)['tour'])['fitness'])
else:
d = m.dist(inst_a, inst_b)
res.append(
solver.solve(
inst_b,
seed=d['pi'][inst_a.optimal_tour])['fitness'])
with open(ofn, 'a') as f:
print(keys[i],
k,
inst_b.eval(inst_b.optimal_tour)[0],
file=f,
end=' ')
for r in res:
print(r, file=f, end=' ')
print(file=f)
def main0():
ia = TSP(n)
ib = TSP(n)
metric = Map_Dist()
res_ident = metric.dist(ia, ia)
res_indep = metric.dist(ia, ib)
print('ident:', res_ident['dist'], res_ident['pi'])
print('indep:', res_indep['dist'], res_indep['pi'])
def main0():
insts = [TSP(n) for _ in range(10)]
for k in kernels:
metric = Kernel_Dist(k)
print(k)
for i in range(0, 10, 2):
res = metric.dist(insts[i], insts[i + 1])
print('\t', res['dist'])
print()
def acc_exp_main(data_dir):
m = 3
metric = Map_Dist()
ofn = 'dists.csv'
with open(os.path.join(data_dir, ofn), 'w') as f:
print('Inst_a,Inst_b,Distance,Baseline', file=f)
fns_ = os.listdir(data_dir)
fns = list(filter(lambda s: s.endswith('.pkl'), fns_))
data = {}
for fn in fns:
with open(os.path.join(data_dir, fn), 'rb') as f:
d = pickle.load(f)
keys = list(d.keys())
i = np.argmin([len(k) for k in keys])
data[keys[i]] = d[keys[i]]
# data.update()
idx = list(data.keys())
n = len(idx)
met_dist_mat = np.zeros((n, n))
har_dist_mat = np.zeros((n, n))
print('#instances:', n)
for i in range(n):
for j in range(n):
inst_i = TSP(data[idx[i]])
inst_j = TSP(data[idx[j]])
met_res = []
har_res = []
for k in range(m):
met_res.append(metric.dist(inst_i, inst_j)['dist'])
har_res.append(abs(inst_i.hardness_est() - inst_j.hardness_est()))
met_dist_mat[i][j] = np.mean(met_res)
har_dist_mat[i][j] = np.mean(har_res)
s = (idx[i], idx[j], str(met_dist_mat[i][j]), str(har_dist_mat[i][j]))
print(s)
with open(os.path.join(data_dir, ofn), 'a') as f:
print(','.join(s), file=f)
def tune_init_exp_main(data_dir):
metrics = [
Map_Dist(sample_size=100),
Map_Dist(sample_size=150),
Map_Dist(sample_size=200),
Map_Dist(sample_size=250),
Map_Dist(sample_size=300)
]
lbls = [100, 150, 200, 250, 300]
solver = GA_Solver()
ofn = 'tune_X.txt'
with open(ofn, 'w') as f:
print('name best', file=f, end=' ')
for l in lbls:
print(l, file=f, end=' ')
print(file=f)
fns_ = os.listdir(data_dir)
fns = list(filter(lambda s: s.endswith('.pkl'), fns_))
for fn in fns:
with open(os.path.join(data_dir, fn), 'rb') as f:
data = pickle.load(f)
keys = list(data.keys())
i = np.argmin([len(k) for k in keys])
inst_a = TSP(data[keys[i]])
inst_b = TSP(data[keys[i] + '_r9_#0'])
res = []
print(keys[i])
for m in metrics:
d = m.dist(inst_a, inst_b)
res.append(
solver.solve(inst_b,
seed=d['pi'][inst_a.optimal_tour])['fitness'])
with open(ofn, 'a') as f:
print(keys[i],
inst_b.eval(inst_b.optimal_tour)[0],
file=f,
end=' ')
for r in res:
print(r, file=f, end=' ')
print(file=f)
def read_inst(fn, dir=None):
if dir is not None:
fn = os.path.join(dir, fn)
points = []
with open(fn) as f:
for line in f:
point = line.rstrip().split()
points.append(point)
return TSP(np.array(points, dtype=float))
def travel_from_seed(inst_0, c):
inst = TSP(inst_0)
n = len(inst)
r = np.mean(cdist(np.zeros((1, 2)), inst)) / np.sqrt(n) * c
alpha = np.random.random(n) * np.pi * 2
delta = np.zeros((n, 2))
delta[:, 0] = np.sin(alpha) * r
delta[:, 1] = np.cos(alpha) * r
inst += delta
return inst
def main1():
ident = []
indep = []
print('prob size', n)
metric = Map_Dist()
for i in range(30):
print('\r', i, end='')
ia = TSP(n)
ib = TSP(n)
id_res = metric.dist(ia, ia)
in_res = metric.dist(ia, ib)
ident.append(id_res['dist'])
indep.append(in_res['dist'])
print('\r', end='')
print('identical set mean', np.mean(ident))
print('identical set std', np.std(ident))
print('independent set mean', np.mean(indep))
print('independent set std', np.std(indep))
print('p =', ttest_ind(ident, indep)[1])
def tsp_instance(n, phi=None):
rf_x, rf_y = random_func(phi)
inst = np.zeros((n, 2))
inst[:, 0] = rf_x(size=n)
inst[:, 1] = rf_y(size=n)
while True:
idx = (cdist(np.zeros((1, 2)), inst) > 1)[0]
m = np.sum(idx)
if m == 0:
return TSP(inst)
inst[idx, 0] = rf_x(size=m)
inst[idx, 1] = rf_y(size=m)
def harden_seed(inst_0, max_iter=100, num_candi=10):
inst = TSP(inst_0)
for i in range(max_iter):
inst_new = inst
fit_new = inst_new.hardness_est()
for j in range(num_candi):
inst_x = resample_from_seed(inst, 0.1, 0.5)
fit_x = inst_x.hardness_est()
if fit_new > fit_x:
inst_new = inst_x
fit_new = fit_x
print(fit_new)
inst = inst_new
return inst
def resample_from_seed(inst_0, c, std=0.0):
inst = TSP(inst_0)
n = len(inst)
m = int(c * n)
idx = np.random.choice(n, m, replace=False)
if std:
inst[idx] = np.random.normal(inst[idx], scale=std)
# bounce out of scope points back
idx = (cdist(np.zeros((1, 2)), inst) > 1)[0]
points = np.array(inst)[idx]
norms = np.linalg.norm(points, axis=1)
inst[idx, 0] = points[:, 0] / (norms**2)
inst[idx, 1] = points[:, 1] / (norms**2)
else:
inst[idx] = tsp_instance(m)
return inst
0.2,
self._max_iter,
verbose=False,
stats=stats,
halloffame=hof)
# population, logbook = algorithms.eaSimple(population, toolbox, 0.7, 0.2, self._max_iter, verbose=False)
tour = np.array(hof[0])
if eval is None:
return {
'tour': tour,
'fitness': inst.eval(tour)[0],
'pop': population,
'iter_avg': [iter['avg'] for iter in logbook]
}
else:
return {
'tour': tour,
'fitness': eval(tour)[0],
'pop': population,
'iter_avg': [iter['avg'] for iter in logbook]
}
if __name__ == '__main__':
tsp = TSP(50)
slr = GA_Solver()
res = slr.solve(tsp)
print(list(res['iter_avg']))
def contr_exp_main(data_dir):
metrics = [
Map_Dist(),
Geo_Dist(),
Geo_Dist(embedding='deepwalk'),
Geo_Dist(embedding='node2vec'),
# Geo_Dist(embedding='line'),
# Geo_Dist(embedding='sdne'),
Geo_Dist(embedding='struc2vec'),
Abstract_Dist(),
Kernel_Dist('edge_hist_gauss'),
Kernel_Dist('edge_hist'),
Kernel_Dist('random_walk_exp'),
Kernel_Dist('random_walk_geo'),
Kernel_Dist('vertex_edge_hist_gauss'),
Kernel_Dist('vertex_edge_hist'),
Kernel_Dist('vertex_vertex_edge_hist'),
]
ofn = 'contr_0.txt'
with open(ofn, 'w') as f:
print('inst_a inst_b inst_c', file=f, end=' ')
for l in [m.lbl for m in metrics]:
print(l, file=f, end=' ')
print(file=f)
fns_ = os.listdir(data_dir)
fns = list(filter(lambda s: s.endswith('.pkl'), fns_))
data = dict()
for fn in fns:
with open(os.path.join(data_dir, fn), 'rb') as f:
data_ = pickle.load(f)
data.update(data_)
keys = list(data.keys())
# print(keys)
for _ in range(30):
for cat in ['circuit', 'city', 'gaussian', 'uniform']:
ks = list()
ks.append(random.choice([k for k in keys if k.startswith(cat)]))
ks.append(
random.choice([
k for k in keys if k.startswith(cat)
and k.split('_')[1] == ks[0].split('_')[1]
]))
ks.append(
random.choice([
k for k in keys if k.startswith(cat)
and k.split('_')[1] != ks[0].split('_')[1]
]))
ks.append(random.choice([k for k in keys
if not k.startswith(cat)]))
print(ks)
insts = [TSP(data[k]) for k in ks]
for triples in [[(0, 1), (0, 3), (1, 3)], [(0, 2), (0, 3),
(2, 3)]]:
res = []
for m in metrics:
for i, j in triples:
if m.lbl.startswith('kernel'):
res.append(1.0 /
m.dist(insts[i], insts[j])['dist'])
else:
res.append(m.dist(insts[i], insts[j])['dist'])
with open(ofn, 'a') as f:
for i, k in enumerate(ks):
if (0, 1) in triples and i == 2:
continue
if (0, 2) in triples and i == 1:
continue
print(k, file=f, end=' ')
for r in res:
print(r, file=f, end=' ')
print(file=f)
def retr_exp_main(data_dir):
metrics = [
Map_Dist(),
Geo_Dist(),
Geo_Dist(embedding='deepwalk'),
Geo_Dist(embedding='node2vec'),
# Geo_Dist(embedding='line'),
# Geo_Dist(embedding='sdne'),
Geo_Dist(embedding='struc2vec'),
Abstract_Dist(),
Kernel_Dist('edge_hist_gauss'),
Kernel_Dist('edge_hist'),
Kernel_Dist('random_walk_exp'),
Kernel_Dist('random_walk_geo'),
Kernel_Dist('vertex_edge_hist_gauss'),
Kernel_Dist('vertex_edge_hist'),
Kernel_Dist('vertex_vertex_edge_hist'),
DMST_Heuristic(),
FI_Heuristic(),
# Gr_Heuristic(),
NI_Heuristic(),
NN_Heuristic()
]
solver = GA_Solver()
ofn = 'retr_0.txt'
with open(ofn, 'w') as f:
print('query metric candidate dist fitness', file=f)
fns_ = os.listdir(data_dir)
fns = list(filter(lambda s: s.endswith('.pkl'), fns_))
data = dict()
for fn in fns:
with open(os.path.join(data_dir, fn), 'rb') as f:
data_ = pickle.load(f)
data.update(data_)
keys = list(data.keys())
queries = list(filter(lambda k: k[-2] != '#', keys))
random.shuffle(queries)
for q in queries:
# candidates = list(filter(lambda k: k.startswith(q), keys))
# candidates.remove(q)
# candidates.extend(queries)
# candidates.remove(q)
candidates = list(queries)
candidates.remove(q)
inst_b = TSP(data[q])
for m in metrics:
print(q, m.lbl)
if m.lbl.startswith('heuristic'):
fit = solver.solve(inst_b,
seed=m.solve(inst_b)['tour'])['fitness']
with open(ofn, 'a') as f:
print(q, m.lbl, None, None, fit, file=f)
continue
seeds = []
for c in candidates:
inst_a = TSP(data[c])
d = m.dist(inst_a, inst_b)
val = d['dist']
if m.lbl.startswith('kernel'):
val = 1.0 / val if val != 0.0 else np.inf
seed = d['pi'][inst_a.optimal_tour]
seeds.append((val, seed))
fit = solver.solve(inst_b, seed=seed)['fitness']
with open(ofn, 'a') as f:
print(q, m.lbl, c, val, fit, file=f)
if m.lbl == 'mapping':
seeds = [s[1] for s in sorted(seeds, key=lambda x: x[0])]
for k in [3, 5, 10]:
fit = solver.solve(inst_b, seed=seeds)['fitness']
with open(ofn, 'a') as f:
print(q, m.lbl, 'top_k', k, fit, file=f)
def sub_sample(inst_0, n):
idx = np.sort(np.random.choice(len(inst_0), n, replace=False))
return TSP(inst_0[idx])
model = DeepWalk(G, walk_length=10, num_walks=80, workers=1)
model.train(window_size=5, iter=3)
elif self._embedding == 'node2vec':
model = Node2Vec(G, walk_length=10, num_walks=80, p=0.25, q=4, workers=1) # init model
model.train(window_size=5, iter=3) # train model
elif self._embedding == 'line':
model = LINE(G, embedding_size=128, order='second') # init model,order can be ['first','second','all']
model.train(batch_size=1024, epochs=50, verbose=2) # train model
elif self._embedding == 'sdne':
model = SDNE(G, hidden_size=[256, 128]) # init model
model.train(batch_size=3000, epochs=40, verbose=2) # train model
elif self._embedding == 'struc2vec':
model = Struc2Vec(G, 10, 80, workers=4, verbose=40, ) # init model
model.train(window_size=5, iter=3) # train model
else:
return self._normalise(inst)
ebds = model.get_embeddings()
coords = []
for i in range(inst.n):
coords.append(ebds[str(i)])
return np.array(coords)
if __name__ == '__main__':
# metric = Geo_Dist(embedding='deepwalk')
metric = Geo_Dist()
inst_a = TSP(10)
inst_b = TSP(10)
print(metric.dist(inst_a, inst_b))
def exp_main(data_name, out_name):
with open(data_name + '.pkl', 'rb') as f:
data = pickle.load(f)
insts = data.keys()
seeds = list(filter(lambda x: '#' not in x, insts))
metric = Map_Dist()
solver = GA_Solver()
heuristics = [NN_Heuristic(), Gr_Heuristic(), NI_Heuristic(), FI_Heuristic(), DMST_Heuristic()]
res = dict()
m = 5
for ia in seeds:
res_ia = dict()
print(ia)
inst_a = TSP(data[ia])
print('other inst init')
for ib in seeds:
if ia == ib:
continue
inst_b = TSP(data[ib])
pi = metric.dist(inst_b, inst_a)['pi']
init = pi[inst_b.optimal_tour]
iter_avg = []
for k in range(m):
sol = solver.solve(inst_a, seed=init)
iter_avg.append(sol['iter_avg'])
iter_avg = np.array(iter_avg)
res_ia[ib] = (init, iter_avg.sum(axis=0))
print('closest inst init')
for ib in insts:
if ia not in ib or ('r1' not in ib and 't1' not in ib):
continue
inst_b = TSP(data[ib])
pi = metric.dist(inst_b, inst_a)['pi']
init = pi[inst_b.optimal_tour]
iter_avg = []
for k in range(m):
sol = solver.solve(inst_a, seed=init)
iter_avg.append(sol['iter_avg'])
iter_avg = np.array(iter_avg)
res_ia[ib] = (init, iter_avg.sum(axis=0))
print('heuristic init')
for heuristic in heuristics:
init = heuristic.solve(inst_a)['tour']
iter_avg = []
for k in range(m):
sol = solver.solve(inst_a, seed=init)
iter_avg.append(sol['iter_avg'])
iter_avg = np.array(iter_avg)
res_ia[heuristic.__class__.__name__] = (init, iter_avg.sum(axis=0))
print('random init')
iter_avg = []
for k in range(m):
sol = solver.solve(inst_a)
iter_avg.append(sol['iter_avg'])
iter_avg = np.array(iter_avg)
res_ia['random'] = (init, iter_avg.sum(axis=0))
res[ia] = res_ia
with open(out_name, 'wb') as f:
pickle.dump(res, f, protocol=pickle.HIGHEST_PROTOCOL)
u, v = filter(lambda x: neigh[x][1] == -1, range(inst.n))
neigh[u][1] = v
neigh[v][1] = u
tour = []
in_tour = [False] * inst.n
tour.append(0)
in_tour[0] = True
while True:
u = tour[-1]
v = neigh[u][0]
if not in_tour[v]:
tour.append(v)
in_tour[v] = True
continue
v = neigh[u][1]
if not in_tour[v]:
tour.append(v)
in_tour[v] = True
continue
break
tour = np.array(tour)
return {'tour': tour, 'fitness': inst.eval(tour)}
if __name__ == '__main__':
tsp = TSP(10)
slr = Gr_Heuristic()
sol = slr.solve(tsp)