Source code for bob.bio.gmm.algorithm.GMM

#!/usr/bin/env python
# vim: set fileencoding=utf-8 :
# Manuel Guenther <Manuel.Guenther@idiap.ch>


import logging

from multiprocessing.pool import ThreadPool

import numpy

import bob.core
import bob.io.base
import bob.learn.em

from bob.bio.base.algorithm import Algorithm

logger = logging.getLogger("bob.bio.gmm")


class GMM(Algorithm):
    """Algorithm for computing Universal Background Models and Gaussian Mixture Models of the features.
    Features must be normalized to zero mean and unit standard deviation."""

    def __init__(
        self,
        # parameters for the GMM
        number_of_gaussians,
        # parameters of UBM training
        kmeans_training_iterations=25,  # Maximum number of iterations for K-Means
        gmm_training_iterations=25,  # Maximum number of iterations for ML GMM Training
        training_threshold=5e-4,  # Threshold to end the ML training
        variance_threshold=5e-4,  # Minimum value that a variance can reach
        update_weights=True,
        update_means=True,
        update_variances=True,
        # parameters of the GMM enrollment
        relevance_factor=4,  # Relevance factor as described in Reynolds paper
        gmm_enroll_iterations=1,  # Number of iterations for the enrollment phase
        responsibility_threshold=0,  # If set, the weight of a particular Gaussian will at least be greater than this threshold. In the case the real weight is lower, the prior mean value will be used to estimate the current mean and variance.
        INIT_SEED=5489,
        # scoring
        scoring_function=bob.learn.em.linear_scoring,
        n_threads=None,
    ):
        """Initializes the local UBM-GMM tool chain with the given file selector object"""

        # call base class constructor and register that this tool performs projection
        Algorithm.__init__(
            self,
            performs_projection=True,
            use_projected_features_for_enrollment=False,
            number_of_gaussians=number_of_gaussians,
            kmeans_training_iterations=kmeans_training_iterations,
            gmm_training_iterations=gmm_training_iterations,
            training_threshold=training_threshold,
            variance_threshold=variance_threshold,
            update_weights=update_weights,
            update_means=update_means,
            update_variances=update_variances,
            relevance_factor=relevance_factor,
            gmm_enroll_iterations=gmm_enroll_iterations,
            responsibility_threshold=responsibility_threshold,
            INIT_SEED=INIT_SEED,
            scoring_function=str(scoring_function),
            multiple_model_scoring=None,
            multiple_probe_scoring="average",
        )

        # copy parameters
        self.gaussians = number_of_gaussians
        self.kmeans_training_iterations = kmeans_training_iterations
        self.gmm_training_iterations = gmm_training_iterations
        self.training_threshold = training_threshold
        self.variance_threshold = variance_threshold
        self.update_weights = update_weights
        self.update_means = update_means
        self.update_variances = update_variances
        self.relevance_factor = relevance_factor
        self.gmm_enroll_iterations = gmm_enroll_iterations
        self.init_seed = INIT_SEED
        self.rng = bob.core.random.mt19937(self.init_seed)
        self.responsibility_threshold = responsibility_threshold
        self.scoring_function = scoring_function
        self.n_threads = n_threads
        self.pool = None

        self.ubm = None
        self.kmeans_trainer = bob.learn.em.KMeansTrainer()
        self.ubm_trainer = bob.learn.em.ML_GMMTrainer(
            self.update_means,
            self.update_variances,
            self.update_weights,
            self.responsibility_threshold,
        )

    def _check_feature(self, feature):
        """Checks that the features are appropriate"""
        if (
            not isinstance(feature, numpy.ndarray)
            or feature.ndim != 2
            or feature.dtype != numpy.float64
        ):
            raise ValueError("The given feature is not appropriate")
        if self.ubm is not None and feature.shape[1] != self.ubm.shape[1]:
            raise ValueError(
                "The given feature is expected to have %d elements, but it has %d"
                % (self.ubm.shape[1], feature.shape[1])
            )

    #######################################################
    #                UBM training                         #

[docs] def train_ubm(self, array): logger.debug(" .... Training with %d feature vectors", array.shape[0]) if self.n_threads is not None: self.pool = ThreadPool(self.n_threads) # Computes input size input_size = array.shape[1] # Creates the machines (KMeans and GMM) logger.debug(" .... Creating machines") kmeans = bob.learn.em.KMeansMachine(self.gaussians, input_size) self.ubm = bob.learn.em.GMMMachine(self.gaussians, input_size) # Trains using the KMeansTrainer logger.info(" -> Training K-Means") # Reseting the pseudo random number generator so we can have the same initialization for serial and parallel execution. self.rng = bob.core.random.mt19937(self.init_seed) bob.learn.em.train( self.kmeans_trainer, kmeans, array, self.kmeans_training_iterations, self.training_threshold, rng=self.rng, pool=self.pool, ) variances, weights = kmeans.get_variances_and_weights_for_each_cluster(array) means = kmeans.means # Initializes the GMM self.ubm.means = means self.ubm.variances = variances self.ubm.weights = weights self.ubm.set_variance_thresholds(self.variance_threshold) # Trains the GMM logger.info(" -> Training GMM") # Reseting the pseudo random number generator so we can have the same initialization for serial and parallel execution. self.rng = bob.core.random.mt19937(self.init_seed) bob.learn.em.train( self.ubm_trainer, self.ubm, array, self.gmm_training_iterations, self.training_threshold, rng=self.rng, pool=self.pool, )
[docs] def save_ubm(self, projector_file): """Save projector to file""" # Saves the UBM to file logger.debug(" .... Saving model to file '%s'", projector_file) hdf5 = ( projector_file if isinstance(projector_file, bob.io.base.HDF5File) else bob.io.base.HDF5File(projector_file, "w") ) self.ubm.save(hdf5)
[docs] def train_projector(self, train_features, projector_file): """Computes the Universal Background Model from the training ("world") data""" [self._check_feature(feature) for feature in train_features] logger.info( " -> Training UBM model with %d training files", len(train_features) ) # Loads the data into an array array = numpy.vstack(train_features) self.train_ubm(array) self.save_ubm(projector_file)
####################################################### # GMM training using UBM #
[docs] def load_ubm(self, ubm_file): hdf5file = bob.io.base.HDF5File(ubm_file) # read UBM self.ubm = bob.learn.em.GMMMachine(hdf5file) self.ubm.set_variance_thresholds(self.variance_threshold)
[docs] def load_projector(self, projector_file): """Reads the UBM model from file""" # read UBM self.load_ubm(projector_file) # prepare MAP_GMM_Trainer kwargs = ( dict( mean_var_update_responsibilities_threshold=self.responsibility_threshold ) if self.responsibility_threshold > 0.0 else dict() ) self.enroll_trainer = bob.learn.em.MAP_GMMTrainer( self.ubm, relevance_factor=self.relevance_factor, update_means=True, update_variances=False, **kwargs ) self.rng = bob.core.random.mt19937(self.init_seed)
[docs] def project_ubm(self, array): logger.debug(" .... Projecting %d feature vectors" % array.shape[0]) # Accumulates statistics gmm_stats = bob.learn.em.GMMStats(self.ubm.shape[0], self.ubm.shape[1]) self.ubm.acc_statistics(array, gmm_stats) # return the resulting statistics return gmm_stats
[docs] def project(self, feature): """Computes GMM statistics against a UBM, given an input 2D numpy.ndarray of feature vectors""" self._check_feature(feature) return self.project_ubm(feature)
[docs] def read_gmm_stats(self, gmm_stats_file): """Reads GMM stats from file.""" return bob.learn.em.GMMStats(bob.io.base.HDF5File(gmm_stats_file))
[docs] def read_feature(self, feature_file): """Read the type of features that we require, namely GMM_Stats""" return self.read_gmm_stats(feature_file)
[docs] def enroll_gmm(self, array): logger.debug(" .... Enrolling with %d feature vectors", array.shape[0]) gmm = bob.learn.em.GMMMachine(self.ubm) gmm.set_variance_thresholds(self.variance_threshold) bob.learn.em.train( self.enroll_trainer, gmm, array, self.gmm_enroll_iterations, self.training_threshold, rng=self.rng, pool=self.pool, ) return gmm
[docs] def enroll(self, feature_arrays): """Enrolls a GMM using MAP adaptation, given a list of 2D numpy.ndarray's of feature vectors""" [self._check_feature(feature) for feature in feature_arrays] array = numpy.vstack(feature_arrays) # Use the array to train a GMM and return it return self.enroll_gmm(array)
###################################################### # Feature comparison #
[docs] def read_model(self, model_file): """Reads the model, which is a GMM machine""" return bob.learn.em.GMMMachine(bob.io.base.HDF5File(model_file))
[docs] def score(self, model, probe): """Computes the score for the given model and the given probe using the scoring function from the config file""" assert isinstance(model, bob.learn.em.GMMMachine) assert isinstance(probe, bob.learn.em.GMMStats) return self.scoring_function( [model], self.ubm, [probe], [], frame_length_normalisation=True )[0][0]
[docs] def score_for_multiple_probes(self, model, probes): """This function computes the score between the given model and several given probe files.""" assert isinstance(model, bob.learn.em.GMMMachine) for probe in probes: assert isinstance(probe, bob.learn.em.GMMStats) # logger.warn("Please verify that this function is correct") return self.probe_fusion_function( self.scoring_function( [model], self.ubm, probes, [], frame_length_normalisation=True ) )
class GMMRegular(GMM): """Algorithm for computing Universal Background Models and Gaussian Mixture Models of the features""" def __init__(self, **kwargs): """Initializes the local UBM-GMM tool chain with the given file selector object""" # logger.warn("This class must be checked. Please verify that I didn't do any mistake here. I had to rename 'train_projector' into a 'train_enroller'!") # initialize the UBMGMM base class GMM.__init__(self, **kwargs) # register a different set of functions in the Tool base class Algorithm.__init__( self, requires_enroller_training=True, performs_projection=False ) ####################################################### # UBM training #
[docs] def train_enroller(self, train_features, enroller_file): """Computes the Universal Background Model from the training ("world") data""" train_features = [feature for client in train_features for feature in client] return self.train_projector(train_features, enroller_file)
####################################################### # GMM training using UBM #
[docs] def load_enroller(self, enroller_file): """Reads the UBM model from file""" return self.load_projector(enroller_file)
###################################################### # Feature comparison #
[docs] def score(self, model, probe): """Computes the score for the given model and the given probe. The score are Log-Likelihood. Therefore, the log of the likelihood ratio is obtained by computing the following difference.""" assert isinstance(model, bob.learn.em.GMMMachine) self._check_feature(probe) score = sum( model.log_likelihood(probe[i, :]) - self.ubm.log_likelihood(probe[i, :]) for i in range(probe.shape[0]) ) return score / probe.shape[0]
[docs] def score_for_multiple_probes(self, model, probes): raise NotImplementedError("Implement Me!")