def test_mogsm(self):
mcgsm = MCGSM(
dim_in=0,
dim_out=3,
num_components=2,
num_scales=2,
num_features=0)
p0 = 0.3
p1 = 0.7
N = 20000
m0 = array([[2], [0], [0]])
m1 = array([[0], [2], [1]])
C0 = cov(randn(mcgsm.dim_out, mcgsm.dim_out**2))
C1 = cov(randn(mcgsm.dim_out, mcgsm.dim_out**2))
input = zeros([0, N])
output = hstack([
dot(cholesky(C0), randn(mcgsm.dim_out, round(p0 * N))) + m0,
dot(cholesky(C1), randn(mcgsm.dim_out, round(p1 * N))) + m1]) * (rand(1, N) + 0.5)
mcgsm.train(input, output, parameters={
'verbosity': 0,
'max_iter': 10,
'train_means': True})
mogsm = MoGSM(3, 2, 2)
# translate parameters from MCGSM to MoGSM
mogsm.priors = sum(exp(mcgsm.priors), 1) / sum(exp(mcgsm.priors))
for k in range(mogsm.num_components):
mogsm[k].mean = mcgsm.means[:, k]
mogsm[k].covariance = inv(dot(mcgsm.cholesky_factors[k], mcgsm.cholesky_factors[k].T))
mogsm[k].scales = exp(mcgsm.scales[k, :])
mogsm[k].priors = exp(mcgsm.priors[k, :]) / sum(exp(mcgsm.priors[k, :]))
self.assertAlmostEqual(mcgsm.evaluate(input, output), mogsm.evaluate(output), 5)
mogsm_samples = mogsm.sample(N)
mcgsm_samples = mcgsm.sample(input)
# generated samples should have the same distribution
for i in range(mogsm.dim):
self.assertTrue(ks_2samp(mogsm_samples[i], mcgsm_samples[0]) > 0.0001)
self.assertTrue(ks_2samp(mogsm_samples[i], mcgsm_samples[1]) > 0.0001)
self.assertTrue(ks_2samp(mogsm_samples[i], mcgsm_samples[2]) > 0.0001)
posterior = mcgsm.posterior(input, mcgsm_samples)
# average posterior should correspond to prior
for k in range(mogsm.num_components):
self.assertLess(abs(1 - mean(posterior[k]) / mogsm.priors[k]), 0.1)
def test_pickle(self):
models = [
Mixture(dim=5),
MoGSM(dim=3, num_components=4, num_scales=7)]
for _ in range(3):
models[0].add_component(GSM(models[0].dim, 7))
for model0 in models:
tmp_file = mkstemp()[1]
# store model
with open(tmp_file, 'w') as handle:
dump({'model': model0}, handle)
# load model
with open(tmp_file) as handle:
model1 = load(handle)['model']
# make sure parameters haven't changed
self.assertEqual(model0.dim, model1.dim)
self.assertEqual(model0.num_components, model1.num_components)
for k in range(model0.num_components):
self.assertLess(max(abs(model0[k].scales - model0[k].scales)), 1e-10)
self.assertLess(max(abs(model0[k].priors - model1[k].priors)), 1e-10)
self.assertLess(max(abs(model0[k].mean - model1[k].mean)), 1e-10)
self.assertLess(max(abs(model0[k].covariance - model1[k].covariance)), 1e-10)
def test_basics(self): model = MoGSM(1, 4, 1) model.priors = arange(model.num_components) + 1. model.priors = model.priors / sum(model.priors) for k in range(model.num_components): model[k].mean = [[k]] model[k].scales = [[1000.]] n = 1000 samples = asarray(model.sample(n) + .5, dtype=int) for k in range(model.num_components): p = model.priors.ravel()[k] x = sum(samples == k) c = binom.cdf(x, n, p) self.assertGreater(c, 1e-5) self.assertGreater(1. - c, 1e-5)
def main(argv):
parser = ArgumentParser(argv[0], description=__doc__)
parser.add_argument('--data_train', '-d', type=str, default='data/BSDS300_train.mat')
parser.add_argument('--data_test', '-t', type=str, default='data/BSDS300_test.mat')
parser.add_argument('--patch_size', '-p', type=int, default=8)
args = parser.parse_args(argv[1:])
A = eye(args.patch_size) - 1. / args.patch_size**2
A[-1] = 1. / args.patch_size**2
logjacobian = slogdet(A)[1]
data_train = loadmat(args.data_train)['data']
data_test = loadmat(args.data_test)['data']
dc_train = dc_component(data_train, args.patch_size)
dc_test = dc_component(data_test, args.patch_size)
h_train, bins = histogram(dc_train, 60, density=True)
h_test, bins = histogram(dc_test, bins, density=False)
model = MoGSM(dim=1, num_components=16, num_scales=4)
model.train(dc_train, parameters={'max_iter': 100})
figure(sans_serif=True)
t = linspace(0, 1, 100)
hist(dc_train.ravel(), 100, density=True)
plot(t, exp(model.loglikelihood(t[None]).ravel()), 'k', line_width=2)
axis(width=5, height=5)
savefig('dc_fit.tex')
loglik = mean(model.loglikelihood(dc_test))
print 'Add these two numbers to your results:'
print 'Log-likelihood (MoGSM): {0:.4f} [nat]'.format(loglik)
print 'Log-likelihood (histogram): {0:.4f} [nat]'.format(sum(h_test * log(h_train)) / sum(h_test))
print 'Log-Jacobian: {0:.4f} [nat]'.format(logjacobian)
return 0
def main(argv):
parser = ArgumentParser(argv[0], description=__doc__)
parser.add_argument('--data_train',
'-d',
type=str,
default='data/BSDS300_train.mat')
parser.add_argument('--data_test',
'-t',
type=str,
default='data/BSDS300_test.mat')
parser.add_argument('--patch_size', '-p', type=int, default=8)
args = parser.parse_args(argv[1:])
A = eye(args.patch_size) - 1. / args.patch_size**2
A[-1] = 1. / args.patch_size**2
logjacobian = slogdet(A)[1]
data_train = loadmat(args.data_train)['data']
data_test = loadmat(args.data_test)['data']
dc_train = dc_component(data_train, args.patch_size)
dc_test = dc_component(data_test, args.patch_size)
h_train, bins = histogram(dc_train, 60, density=True)
h_test, bins = histogram(dc_test, bins, density=False)
model = MoGSM(dim=1, num_components=16, num_scales=4)
model.train(dc_train, parameters={'max_iter': 100})
figure(sans_serif=True)
t = linspace(0, 1, 100)
hist(dc_train.ravel(), 100, density=True)
plot(t, exp(model.loglikelihood(t[None]).ravel()), 'k', line_width=2)
axis(width=5, height=5)
savefig('dc_fit.tex')
loglik = mean(model.loglikelihood(dc_test))
print 'Add these two numbers to your results:'
print 'Log-likelihood (MoGSM): {0:.4f} [nat]'.format(loglik)
print 'Log-likelihood (histogram): {0:.4f} [nat]'.format(
sum(h_test * log(h_train)) / sum(h_test))
print 'Log-Jacobian: {0:.4f} [nat]'.format(logjacobian)
return 0
def test_mogsm(self):
mcgsm = MCGSM(dim_in=0,
dim_out=3,
num_components=2,
num_scales=2,
num_features=0)
p0 = 0.3
p1 = 0.7
N = 20000
m0 = array([[2], [0], [0]])
m1 = array([[0], [2], [1]])
C0 = cov(randn(mcgsm.dim_out, mcgsm.dim_out**2))
C1 = cov(randn(mcgsm.dim_out, mcgsm.dim_out**2))
input = zeros([0, N])
output = hstack([
dot(cholesky(C0), randn(mcgsm.dim_out, round(p0 * N))) + m0,
dot(cholesky(C1), randn(mcgsm.dim_out, round(p1 * N))) + m1
]) * (rand(1, N) + 0.5)
mcgsm.train(input,
output,
parameters={
'verbosity': 0,
'max_iter': 10,
'train_means': True
})
mogsm = MoGSM(3, 2, 2)
# translate parameters from MCGSM to MoGSM
mogsm.priors = sum(exp(mcgsm.priors), 1) / sum(exp(mcgsm.priors))
for k in range(mogsm.num_components):
mogsm[k].mean = mcgsm.means[:, k]
mogsm[k].covariance = inv(
dot(mcgsm.cholesky_factors[k], mcgsm.cholesky_factors[k].T))
mogsm[k].scales = exp(mcgsm.scales[k, :])
mogsm[k].priors = exp(mcgsm.priors[k, :]) / sum(
exp(mcgsm.priors[k, :]))
self.assertAlmostEqual(mcgsm.evaluate(input, output),
mogsm.evaluate(output), 5)
mogsm_samples = mogsm.sample(N)
mcgsm_samples = mcgsm.sample(input)
# generated samples should have the same distribution
for i in range(mogsm.dim):
self.assertTrue(
ks_2samp(mogsm_samples[i], mcgsm_samples[0]) > 0.0001)
self.assertTrue(
ks_2samp(mogsm_samples[i], mcgsm_samples[1]) > 0.0001)
self.assertTrue(
ks_2samp(mogsm_samples[i], mcgsm_samples[2]) > 0.0001)
posterior = mcgsm.posterior(input, mcgsm_samples)
# average posterior should correspond to prior
for k in range(mogsm.num_components):
self.assertLess(abs(1 - mean(posterior[k]) / mogsm.priors[k]), 0.1)
def main(argv):
parser = ArgumentParser(argv[0], description=__doc__)
parser.add_argument('--data',
'-d',
type=str,
default='data/BSDS300_8x8.mat')
parser.add_argument('--num_train', '-N', type=int, default=1000000)
parser.add_argument('--num_valid', '-V', type=int, default=200000)
parser.add_argument('--num_components', '-n', type=int, default=128)
parser.add_argument('--num_scales', '-s', type=int, default=4)
parser.add_argument('--num_features', '-f', type=int, default=48)
parser.add_argument('--train_means', '-M', type=int, default=0)
parser.add_argument('--indices', '-I', type=int, default=[], nargs='+')
parser.add_argument('--initialize', '-i', type=str, default=None)
parser.add_argument('--verbosity', '-v', type=int, default=1)
parser.add_argument('--max_iter', '-m', type=int, default=2000)
args = parser.parse_args(argv[1:])
experiment = Experiment()
data_train = loadmat(args.data)['patches_train']
data_valid = loadmat(args.data)['patches_valid']
if args.initialize:
results = Experiment(args.initialize)
models = results['models']
preconditioners = results['preconditioners']
else:
models = [None] * data_train.shape[1]
preconditioners = [None] * data_train.shape[1]
def preprocess(data, i, N):
if N > 0 and N < data.shape[0]:
# select subset of data
idx = random_select(N, data.shape[0])
return data[idx, :i].T, data[idx, i][None, :]
return data.T[:i], data.T[[i]]
for i in range(data_train.shape[1]):
if args.indices and i not in args.indices:
# skip this one
continue
print 'Training model {0}/{1}...'.format(i + 1, data_train.shape[1])
inputs_train, outputs_train = preprocess(data_train, i, args.num_train)
inputs_valid, outputs_valid = preprocess(data_valid, i, args.num_valid)
if i > 0:
if preconditioners[i] is None:
preconditioners[i] = WhiteningPreconditioner(
inputs_train, outputs_train)
inputs_train, outputs_train = preconditioners[i](inputs_train,
outputs_train)
inputs_valid, outputs_valid = preconditioners[i](inputs_valid,
outputs_valid)
if models[i] is None:
models[i] = MCGSM(dim_in=i,
dim_out=1,
num_components=args.num_components,
num_features=args.num_features,
num_scales=args.num_scales)
models[i].train(inputs_train,
outputs_train,
inputs_valid,
outputs_valid,
parameters={
'verbosity': 1,
'max_iter': args.max_iter,
'train_means': args.train_means > 0
})
else:
preconditioners[i] = None
if models[i] is None:
models[i] = MoGSM(dim=1, num_components=4, num_scales=8)
models[i].train(outputs_train,
outputs_valid,
parameters={
'verbosity': 1,
'threshold': -1.,
'train_means': 1,
'max_iter': 100
})
experiment['args'] = args
experiment['models'] = models
experiment['preconditioners'] = preconditioners
experiment.save(
'results/BSDS300/snapshots/mcgsm_{0}_{1}.{{0}}.{{1}}.xpck'.format(
i, args.num_components))
if not args.indices:
experiment['args'] = args
experiment['models'] = models
experiment['preconditioners'] = preconditioners
experiment.save('results/BSDS300/mcgsm.{0}.{1}.xpck')
return 0
