GPU/CUDAMat Backend¶

Example Usage¶

Training a GMM:

import ggmm.gpu as ggmm

X = some_module.load_training_data()

# N - training examples
# D - data dimension
# K - number of GMM components
N, D = X.shape
K = 128

ggmm.init()
gmm = ggmm.GMM(K,D)

thresh = 1e-3 # convergence threshold
n_iter = 20 # maximum number of EM iterations
init_params = 'wmc' # initialize weights, means, and covariances

# train GMM
gmm.fit(X, thresh, n_iter, init_params=init_params)

# retrieve parameters from trained GMM
weights = gmm.get_weights()
means = gmm.get_means()
covars = gmm.get_covars()

# compute posteriors of data
posteriors = gmm.compute_posteriors(X)

Reference¶

ggmm.gpu.init(max_ones=262144)¶

Initialize GPU resources.

Parameters:	max_ones : int, optional Allocate enough memory for a sum of up to ‘max_ones’ length

ggmm.gpu.shutdown()¶

Free GPU resources

class ggmm.gpu.GMM(n_components, n_dimensions, covariance_type='diag', min_covar=0.001, verbose=False)¶

Gaussian Mixture Model

Representation of a Gaussian mixture model probability distribution. This class allows for easy evaluation of, sampling from, and maximum-likelihood estimation of the parameters of a GMM distribution.

Initializes parameters such that every mixture component has zero mean and identity covariance.

Parameters:	n_components : int, required Number of mixture components. n_dimensions : int, required Number of data dimensions. covariance_type : string, optional String describing the type of covariance parameters to use. For now, only ‘diag’ is supported. Defaults to ‘diag’. min_covar : float, optional Floor on the diagonal of the covariance matrix to prevent overfitting. Defaults to 1e-3.

Methods

compute_posteriors(X)¶

Predict posterior probability of data under each Gaussian in the model.

Parameters:	X : array-like, shape = [n_samples, n_dimensions]
Returns:	posteriors : array-like, shape = (n_samples, n_components) Returns the probability of the sample for each Gaussian (state) in the model.

fit(X, thresh=0.01, n_iter=100, n_init=1, update_params='wmc', init_params='', random_state=None, verbose=None)¶

Estimate model parameters with the expectation-maximization algorithm.

A initialization step is performed before entering the em algorithm. If you want to avoid this step, set the keyword argument init_params to the empty string ‘’ when creating the GMM object. Likewise, if you would like just to do an initialization, set n_iter=0.

Parameters:

X : array_like, shape (n_samples, n_dimensions) List of ‘n_samples’ data points. Each row corresponds to a single data point. thresh : float, optional Convergence threshold. n_iter : int, optional Number of EM iterations to perform. n_init : int, optional Number of initializations to perform. the best results is kept update_params : string, optional Controls which parameters are updated in the training process. Can contain any combination of ‘w’ for weights, ‘m’ for means, and ‘c’ for covars. Defaults to ‘wmc’. init_params : string, optional Controls which parameters are updated in the initialization process. Can contain any combination of ‘w’ for weights, ‘m’ for means, and ‘c’ for covars. Defaults to ‘wmc’. random_state: numpy.random.RandomState : verbose: bool, optional : Whether to print EM iteration information during training

get_covars()¶

Return current means as numpy array

Returns:	covars : np.ndarray, shape (n_components, n_dimensions) (for now only diagonal covariance matrices are supported)

get_means()¶

Return current means as numpy array

Returns:	means : np.ndarray, shape (n_components, n_dimensions)

get_weights()¶

Return current weight vector as numpy array

Returns:	weights : np.ndarray, shape (n_components,)

predict(X)¶

Predict label for data.

Parameters:	X : array-like, shape = [n_samples, n_dimensions]
Returns:	C : array, shape = (n_samples,)

score(X)¶

Compute the log probability under the model.

Parameters:	X : array_like, shape (n_samples, n_dimensions) List of ‘n_samples’ data points. Each row corresponds to a single data point.
Returns:	logprob_Nx1 : array_like, shape (n_samples,) Log probabilities of each data point in X

score_samples(X, temp_gpu_mem=None)¶

Return the per-sample likelihood of the data under the model.

Compute the log probability of X under the model and return the posterior probability of each mixture component for each element of X.

Parameters:	X: numpy.ndarray, shape (n_samples, n_dimensions) : Array of n_samples data points. Each row corresponds to a single data point.
Returns:	logprob_Nx1 : array_like, shape (n_samples,) Log probabilities of each data point in X. posteriors : array_like, shape (n_samples, n_components) Posterior probability of each mixture component for each sample

set_covars(covars)¶

Set covariance matrices with numpy array

Parameters:	covars: numpy.ndarray, shape (n_components, n_dimensions) : (for now only diagonal covariance matrices are supported)

set_means(means)¶

Set mean vectors with numpy array.

Parameters:	means: numpy.ndarray, shape (n_components, n_dimensions) :

set_weights(weights)¶

Set weight vector with numpy array.

Parameters:	weights: numpy.ndarray, shape (n_components,) :