def test_train_batch_IP_diagonal(self):
training_data = self._training_data()
embed = GaussianEmbedding(10, 5,
covariance_type='diagonal',
energy_type='IP',
mu_max=2.0, sigma_min=0.8, sigma_max=1.2, eta=0.1, Closs=1.0
)
for k in xrange(0, len(training_data), 100):
embed.train_batch(training_data[k:(k+100)])
self._check_results(embed)
def test_train_batch_KL_diagonal(self):
training_data = self._training_data()
embed = GaussianEmbedding(10, 5,
covariance_type='diagonal',
energy_type='KL',
mu_max=2.0, sigma_min=0.8, sigma_max=1.2, eta=0.1, Closs=1.0
)
# diagonal training has more parameters so needs more then one
# epoch to fully learn data
for k in xrange(0, len(training_data), 100):
embed.train_batch(training_data[k:(k+100)])
self._check_results(embed)
def test_train_threads(self):
training_data = self._training_data()
embed = GaussianEmbedding(10, 5,
covariance_type='spherical',
energy_type='KL',
mu_max=2.0, sigma_min=0.8, sigma_max=1.2, eta=0.1, Closs=1.0
)
def iter_pairs():
for k in xrange(0, len(training_data), 100):
yield training_data[k:(k+100)]
embed.train(iter_pairs(), n_workers=4)
self._check_results(embed)
def sample_embed(energy_type='KL', covariance_type='spherical', eta=0.1):
mu = np.array([
[0.0, 0.0],
[1.0, -1.25],
[-0.1, -0.4],
[1.2, -0.3],
[0.5, 0.5],
[-0.55, -0.75]
], dtype=DTYPE)
if covariance_type == 'spherical':
sigma = np.array([
[1.0],
[5.0],
[0.8],
[0.4],
[1.5],
[1.4]
], dtype=DTYPE)
elif covariance_type == 'diagonal':
sigma = np.array([
[1.0, 0.1],
[5.0, 5.5],
[0.8, 1.1],
[0.9, 1.9],
[0.65, 0.9],
[1.5, 1.55]
], dtype=DTYPE)
return GaussianEmbedding(3, size=2,
covariance_type=covariance_type,
energy_type=energy_type,
mu=mu, sigma=sigma, eta=eta
)
def test_eta_multiple(self):
expected = {
'mu': 0.1,
'mu_min': 0.001,
'sigma': 0.05,
'sigma_min': 0.000005
}
embed = GaussianEmbedding(10, 5, eta=expected)
actual = embed.eta
for k, v in list(expected.items()):
self.assertAlmostEqual(actual[k], v)
def test_eta_single(self):
embed = GaussianEmbedding(10, 5, eta=0.55)
expected = {
'mu': 0.55,
'mu_min': 0.0,
'sigma': 0.55,
'sigma_min': 0.0
}
actual = embed.eta
for k, v in expected.items():
self.assertAlmostEqual(actual[k], v)
