def test_emd_equivalence(M1, M2, norm, R):
gdim = 2
events1 = np.random.rand(nev, M1, gdim + 1)
events2 = np.random.rand(nev, M2, gdim + 1)
# test two different sets
emds1 = np.zeros((nev, nev))
for i, ev1 in enumerate(events1):
for j, ev2 in enumerate(events2):
emds1[i, j] = emd.emd(ev1, ev2, R=R, norm=norm, gdim=gdim)
emds2 = emd.emds(events1,
events2,
R=R,
norm=norm,
verbose=0,
n_jobs=1,
gdim=gdim)
assert epsilon_diff(emds1, emds2, 10**-12)
# test same set
emds1 = np.zeros((nev, nev))
for i, ev1 in enumerate(events1):
for j in range(i):
emds1[i, j] = emd.emd(ev1, events1[j], R=R, norm=norm, gdim=gdim)
emds1 += emds1.T
emds2 = emd.emds(events1, R=R, norm=norm, verbose=0, n_jobs=1, gdim=gdim)
assert epsilon_diff(emds1, emds2, 10**-12)
def test_periodic_phi(gdim, M):
events = np.random.rand(nev, M, 1 + gdim)
for phi_col in range(1, gdim + 1):
emds1 = emd.emds(events, R=1.0, gdim=gdim, n_jobs=1, verbose=0)
events_c = np.copy(events)
events_c[:, :, phi_col] += 2 * np.pi * np.random.randint(
-10, 10, size=(nev, M))
emds2 = emd.emds(events_c,
R=1.0,
gdim=gdim,
periodic_phi=True,
phi_col=phi_col,
n_jobs=1,
verbose=0)
assert epsilon_diff(emds1, emds2, 10**-12)
ev1 = np.random.rand(10, 1 + gdim) * 4 * np.pi
ev2 = np.random.rand(20, 1 + gdim) * 4 * np.pi
thetaw = np.zeros((len(ev1), len(ev2)))
thetar = np.zeros((len(ev1), len(ev2)))
for i, p1 in enumerate(ev1):
for j, p2 in enumerate(ev2):
dw, dr = 0., 0.
for m, (k1, k2) in enumerate(zip(p1, p2)):
if m == 0:
continue
elif m == phi_col:
dw += (k1 - k2)**2
dr += np.min([
abs(k1 - (k2 + 2 * np.pi * n))
for n in range(-3, 3)
])**2
else:
dw += (k1 - k2)**2
dr += (k1 - k2)**2
thetaw[i, j] = np.sqrt(dw)
thetar[i, j] = np.sqrt(dr)
zs1 = np.ascontiguousarray(ev1[:, 0] / np.sum(ev1[:, 0]))
zs2 = np.ascontiguousarray(ev2[:, 0] / np.sum(ev2[:, 0]))
ot_w, ot_r = ot.emd2(zs1, zs2, thetaw), ot.emd2(zs1, zs2, thetar)
ef_w = emd.emd(ev1,
ev2,
norm=True,
gdim=gdim,
periodic_phi=False,
phi_col=phi_col)
ef_r = emd.emd(ev1,
ev2,
norm=True,
gdim=gdim,
periodic_phi=True,
phi_col=phi_col)
assert epsilon_diff(ot_w, ef_w, 10**-14)
assert epsilon_diff(ot_r, ef_r, 10**-14)
def test_gdim(gdim, evdim, M, norm, R):
if R < np.sqrt(gdim) / 2:
pytest.skip('R too small')
events = np.random.rand(nev, M, 1 + evdim)
emds1 = emd.emds(events, gdim=gdim, norm=norm, R=R, n_jobs=1, verbose=0)
emds2 = emd.emds(events[:, :, :1 + gdim],
gdim=100,
norm=norm,
R=R,
n_jobs=1,
verbose=0)
assert epsilon_diff(emds1, emds2, 10**-13)
def test_n_jobs(n_jobs, M, norm, R):
events = np.random.rand(nev, M, 3)
emds1 = np.zeros((nev, nev))
for i, ev1 in enumerate(events):
for j in range(i):
emds1[i, j] = emd.emd(ev1, events[j], R=R, norm=norm)
emds1 += emds1.T
emds2 = emd.emds(events, R=R, norm=norm, verbose=0, n_jobs=n_jobs)
assert epsilon_diff(emds1, emds2, 10**-12)
def calc_cov_mmd(args, X, gen_out, losses, labels=None):
X_rn, mask_real = utils.unnorm_data(
args, X.cpu().detach().numpy()[:args.eval_tot_samples], real=True)
gen_out_rn, mask_gen = utils.unnorm_data(args,
gen_out[:args.eval_tot_samples],
real=False)
# converting into EFP format
X_rn = np.concatenate((np.expand_dims(X_rn[:, :, 2], 2), X_rn[:, :, :2],
np.zeros((X_rn.shape[0], X_rn.shape[1], 1))),
axis=2)
gen_out_rn = np.concatenate(
(np.expand_dims(gen_out_rn[:, :, 2], 2), gen_out_rn[:, :, :2],
np.zeros((gen_out_rn.shape[0], gen_out_rn.shape[1], 1))),
axis=2)
if args.clabels == 1:
abs_labels = (labels[:args.eval_tot_samples, 0] *
args.maxjf[0]).detach().numpy()
logging.info("Calculating coverage and MMD")
covs = []
mmds = []
if args.clabels == 1:
intra_covs = []
intra_mmds = []
for j in range(args.cov_mmd_num_batches):
G_rand_sample = rng.choice(args.eval_tot_samples,
size=args.cov_mmd_num_samples)
X_rand_sample = rng.choice(args.eval_tot_samples,
size=args.cov_mmd_num_samples)
Gsample = gen_out_rn[G_rand_sample]
Xsample = X_rn[X_rand_sample]
dists = emds(Gsample, Xsample)
mmds.append(np.mean(np.min(dists, axis=0)))
covs.append(
np.unique(np.argmin(dists, axis=1)).size /
args.cov_mmd_num_samples)
# Intra-W1
if args.clabels == 1:
num_regions = len(pt_regions) - 1
covs_all = []
mmds_all = []
for i in range(num_regions):
cut = (abs_labels > pt_regions[i]) * (abs_labels <
pt_regions[i + 1])
tot_cut = np.sum(cut)
Gcut = gen_out_rn[cut]
Xcut = X_rn[cut]
G_rand_sample = rng.choice(tot_cut,
size=args.cov_mmd_num_samples)
X_rand_sample = rng.choice(tot_cut,
size=args.cov_mmd_num_samples)
Gsample = Gcut[G_rand_sample]
Xsample = Xcut[X_rand_sample]
dists = emds(Gsample, Xsample)
mmds_all += [np.mean(np.min(dists, axis=0))]
covs_all += [
np.unique(np.argmin(dists, axis=1)).size /
args.cov_mmd_num_samples
]
intra_covs.append(covs_all)
intra_mmds.append(mmds_all)
losses['coverage'].append(np.mean(np.array(covs)))
losses['mmd'].append(np.mean(np.array(mmds)))
if args.clabels == 1:
losses['intra_coverage'].append(np.mean(np.array(intra_covs), axis=0))
losses['intra_mmd'].append(np.mean(np.array(intra_mmds), axis=0))
h.heap() importlib.reload(utils) realefp = utils.efp(args, X_rn, mask=mask_real, real=True) genefp = utils.efp(args, gen_out_rn, mask=mask_gen, real=False) save_outputs.plot_jet_feats(args, realjf, genjf, realefp, genefp, 'j', show=True) Gsample = utils.ef_format(gen_out_rn[:10]) Xsample = utils.ef_format(X_rn[:10]) Gsample dists = emds(Gsample, Xsample) dists mmd = np.mean(np.min(dists, axis=0)) cov = np.unique(np.argmin(dists, axis=1)).size / 10 mmd = np.mean(np.min(dists, axis=1)) mmd.append(np.mean(np.min(dists, axis=1))) covs.append()
2), samples_dict[dataset][0][:, :, :2],
np.zeros((samples_dict[dataset][0].shape[0],
samples_dict[dataset][0].shape[1], 1))),
axis=2)
covs = []
mmds = []
for j in range(10):
X_rand_sample = rng.choice(50000, size=100)
X_rand_sample2 = rng.choice(50000, size=100)
Xsample = X_rn[X_rand_sample]
Xsample2 = X_rn[X_rand_sample2]
dists = emds(Xsample, Xsample2)
mmds.append(np.mean(np.min(dists, axis=0)))
covs.append(np.unique(np.argmin(dists, axis=1)).size / 100)
print(f"mmds: {np.mean(mmds)}")
print(f"cov: {np.mean(covs)}")
args_txt = {
'g': 'args/236_g30_dea_no_pos_diffs_graphcnngang_mpgand.txt',
't': 'args/237_t30_lrx2_dea_no_pos_diffs_graphcnngang_mpgand.txt',
'q': 'args/238_q30_lrx05_dea_no_pos_diffs_graphcnngang_mpgand.txt'
}
for dataset in samples_dict.keys():
print(dataset)
