def test():
func = SourceModule(source).get_function('log_diag_mvn_likelihood')
n_samples = 8
n_states = 9
n_features = 33
np.random.seed(42)
sequences = np.random.rand(n_samples, n_features).astype(np.float32)
means = np.random.rand(n_states, n_features).astype(np.float32)
variances = np.random.rand(n_states, n_features).astype(np.float32)
loglikelihoods = np.zeros((n_samples, n_states), dtype=np.float32)
func(cuda.In(sequences),
cuda.In(means),
cuda.In(variances),
cuda.In(np.log(variances)),
np.int32(n_samples),
np.int32(n_states),
np.int32(n_features),
cuda.InOut(loglikelihoods),
block=(64, 1, 1),
grid=(1, 1))
print 'loglikelihoods'
print loglikelihoods
print 'sklearn'
from sklearn.mixture.gmm import _log_multivariate_normal_density_diag
r = _log_multivariate_normal_density_diag(sequences, means, variances)
print r
print np.abs(r - loglikelihoods) < 1e-4
def test_gaussian_likelihood():
np.random.seed(42)
from sklearn.mixture.gmm import _log_multivariate_normal_density_diag
n_states = 4
n_features = 33
n_observations = 10
means = np.random.randn(n_states, n_features).astype(np.float32)
variances = np.random.rand(n_states, n_features).astype(np.float32)
s = np.random.randn(n_observations, n_features).astype(np.float32)
ref = _log_multivariate_normal_density_diag(s, means, variances)
loglikelihoods = np.zeros((n_observations, n_states), dtype=np.float32)
mod.get_function('gaussian_likelihood')(drv.In(s),
drv.In(means),
drv.In(variances),
np.int64(1),
drv.In(np.array([n_observations])),
drv.In(np.array(0)),
np.int64(n_states),
np.int64(n_features),
drv.Out(loglikelihoods),
block=(32, 1, 1),
grid=(1, 1))
print 'reference'
print ref
print 'cuda'
print loglikelihoods
def test_gaussian_likelihood():
np.random.seed(42)
from sklearn.mixture.gmm import _log_multivariate_normal_density_diag
n_states = 4
n_features = 33
n_observations = 10
means = np.random.randn(n_states, n_features).astype(np.float32)
variances = np.random.rand(n_states, n_features).astype(np.float32)
s = np.random.randn(n_observations, n_features).astype(np.float32)
ref = _log_multivariate_normal_density_diag(s, means, variances)
loglikelihoods = np.zeros((n_observations, n_states), dtype=np.float32)
mod.get_function('gaussian_likelihood')(drv.In(s), drv.In(means), drv.In(variances),
np.int64(1), drv.In(np.array([n_observations])), drv.In(np.array(0)),
np.int64(n_states), np.int64(n_features), drv.Out(loglikelihoods),
block=(32,1,1), grid=(1,1))
print 'reference'
print ref
print 'cuda'
print loglikelihoods
def test():
func = SourceModule(source).get_function('log_diag_mvn_likelihood')
n_samples = 8
n_states = 9
n_features = 33
np.random.seed(42)
sequences = np.random.rand(n_samples, n_features).astype(np.float32)
means = np.random.rand(n_states, n_features).astype(np.float32)
variances = np.random.rand(n_states, n_features).astype(np.float32)
loglikelihoods = np.zeros((n_samples, n_states), dtype=np.float32)
func(cuda.In(sequences), cuda.In(means), cuda.In(variances), cuda.In(np.log(variances)),
np.int32(n_samples), np.int32(n_states), np.int32(n_features),
cuda.InOut(loglikelihoods),
block=(64,1,1), grid=(1,1))
print 'loglikelihoods'
print loglikelihoods
print 'sklearn'
from sklearn.mixture.gmm import _log_multivariate_normal_density_diag
r = _log_multivariate_normal_density_diag(sequences, means, variances)
print r
print np.abs(r - loglikelihoods) < 1e-4
