diff --git a/bob/learn/em/mixture/linear_scoring.py b/bob/learn/em/mixture/linear_scoring.py
index 1f04ca777c7a4e362103c850d8fb14a427eaea1f..d7aac713791d95e07d4dbf2ae29a6f3afa56eb3a 100644
--- a/bob/learn/em/mixture/linear_scoring.py
+++ b/bob/learn/em/mixture/linear_scoring.py
@@ -59,7 +59,7 @@ def linear_scoring(
if models_means.ndim == 2:
models_means = models_means[None, :, :]
- if ubm.traier == "map":
+ if ubm.trainer == "map":
ubm = ubm.ubm
if isinstance(test_stats, GMMStats):
diff --git a/bob/learn/em/test/test_gaussian.py b/bob/learn/em/test/test_gaussian.py
deleted file mode 100644
index 45eb5ccf30ad610cc655a73a4c7f99ce435d22ed..0000000000000000000000000000000000000000
--- a/bob/learn/em/test/test_gaussian.py
+++ /dev/null
@@ -1,179 +0,0 @@
-#!/usr/bin/env python
-# vim: set fileencoding=utf-8 :
-# Laurent El Shafey <Laurent.El-Shafey@idiap.ch>
-# Thu Feb 16 16:54:45 2012 +0200
-#
-# Copyright (C) 2011-2013 Idiap Research Institute, Martigny, Switzerland
-
-"""Tests the Gaussian class
-"""
-
-import tempfile
-
-import numpy as np
-
-from h5py import File as HDF5File
-
-from bob.learn.em.mixture import Gaussians
-
-
-def equals(x, y, epsilon):
- return abs(x - y) < epsilon
-
-
-def test_GaussiansObject():
- """Tests a Gaussians object creation and manipulation."""
-
- # Initializes a Gaussians with zero mean and unit variance
- g = Gaussians(means=np.zeros((3,)))
- np.testing.assert_equal(g["means"], np.zeros((1,3)))
- np.testing.assert_equal(g["variances"], np.ones((1,3)))
- np.testing.assert_equal(g.shape, (1,3))
-
- # Set and check mean, variance, variance thresholds
- mean = np.array([[0, 1, 2]], 'float64')
- variance = np.array([[3, 2, 1]], 'float64')
- g["means"] = mean
- g["variances"] = variance
- g["variance_thresholds"] = 0.0005
- np.testing.assert_equal(g["means"], mean)
- np.testing.assert_equal(g["variances"], variance)
- np.testing.assert_equal(g["variance_thresholds"], np.full((1, 3), 0.0005))
-
- # Save and read from file
- filename = str(tempfile.mkstemp(".hdf5")[1])
- g.save(HDF5File(filename, 'w'))
- g_loaded = Gaussians.from_hdf5(HDF5File(filename, "r"))
- assert g == g_loaded
- assert (g != g_loaded) is False
- assert g.is_similar_to(g_loaded)
-
- # Save and read from file using `from_hdf5`
- filename = str(tempfile.mkstemp(".hdf5")[1])
- g.save(hdf5=HDF5File(filename, 'w'))
- g_loaded = Gaussians.from_hdf5(hdf5=HDF5File(filename, "r"))
- assert g == g_loaded
- assert (g != g_loaded) is False
- assert g.is_similar_to(g_loaded)
-
- # Save and load from file using `load` into an existing Gaussians
- filename = str(tempfile.mkstemp(".hdf5")[1])
- g.save(HDF5File(filename, 'w'))
- g_loaded = Gaussians(np.zeros(shape=(3,))) # Dummy fill, same size
- g_loaded.load(HDF5File(filename, "r"))
- assert g == g_loaded
- assert (g != g_loaded) is False
- assert g.is_similar_to(g_loaded)
-
- # Make them different
- g2 = g.copy()
- g["variance_thresholds"] = 0.001
- assert (g == g2) is False
- assert g != g2
- assert not g.is_similar_to(g_loaded)
-
- # Check likelihood computation
- sample1 = np.array([0, 1, 2], 'float64')
- sample2 = np.array([1, 2, 3], 'float64')
- sample3 = np.array([2, 3, 4], 'float64')
- ref1 = -3.652695334228046
- ref2 = -4.569362000894712
- ref3 = -7.319362000894712
- np.testing.assert_almost_equal(g.log_likelihood(sample1), ref1, decimal=10)
- np.testing.assert_almost_equal(g.log_likelihood(sample2), ref2, decimal=10)
- np.testing.assert_almost_equal(g.log_likelihood(sample3), ref3, decimal=10)
-
-
-def test_gaussian_variance_threshold():
- # Creating a Gaussian
- gaussian = Gaussians(means=[1, 2], variances=[1, 1], variance_thresholds=[1e-5, 1e-5])
-
- # Testing variance threshold application
- gaussian["variances"] = np.array([1e-8, 1e-4])
- np.testing.assert_equal(gaussian["variances"], np.array([[1e-5, 1e-4]]))
- gaussian["variance_thresholds"] = np.array([1e-7, 1e-3])
- np.testing.assert_equal(gaussian["variances"], np.array([[1e-5, 1e-3]]))
- gaussian["variances"] = np.array([0, 1e-8])
- np.testing.assert_equal(gaussian["variances"], np.array([[1e-7, 1e-3]]))
- new_gaussian = Gaussians(means=[0,0,0], variances=0, variance_thresholds=1e-5)
- np.testing.assert_equal(new_gaussian["variances"], np.full((1,3), 1e-5))
-
-
-def test_likelihood():
- """Tests the likelihood computation of a simple normal Gaussian."""
- gaussian = Gaussians(means=[0, 0], variances=[1, 1], variance_thresholds=[1e-5, 1e-5])
- log_likelihood = gaussian.log_likelihood(np.array([[0.4, 0.2]], "float64"))
- np.testing.assert_almost_equal(log_likelihood, [[-1.93787706641]], decimal=10)
-
- multi_log_likelihood = gaussian.log_likelihood(
- np.array([[0.4, 0.2], [0.1, 0.3]], "float64")
- )
- expected = np.array([[-1.93787706641, -1.88787706640]])
- np.testing.assert_almost_equal(multi_log_likelihood, expected, decimal=10)
-
- # Default settings
- gaussian_def = Gaussians(means=[1, 2, 3])
- np.testing.assert_equal(gaussian_def["variances"], np.array([[1.0, 1.0, 1.0]]))
- eps = np.finfo(float).eps
- np.testing.assert_equal(
- gaussian_def["variance_thresholds"], np.array([[eps, eps, eps]])
- )
- assert hasattr(gaussian, "log_likelihood")
- log_likelihood = gaussian.log_likelihood(np.array([1.0, 2.0]))
- np.testing.assert_almost_equal(log_likelihood, np.array([[-4.33787706641]]), decimal=10)
-
-
-def test_multiple_gaussian():
- """Tests the capacity to store and work with multiple gaussians."""
- gaussians = Gaussians(
- means=np.array([[0, 0],[3, 3]])
- )
- expected_means = np.array([[0.0, 0.0], [3.0, 3.0]])
- expected_variances = np.array([[1.0, 1.0], [1.0, 1.0]])
- eps = np.finfo(float).eps
- expected_var_thresholds = np.array([[eps, eps], [eps, eps]])
- np.testing.assert_equal(gaussians["means"], expected_means)
- np.testing.assert_equal(gaussians["variances"], expected_variances)
- np.testing.assert_equal(gaussians["variance_thresholds"], expected_var_thresholds)
- test_samples = np.array([[0.0, 0.0], [1.0, 2.0], [3.0, 3.0]])
- log_likelihoods = gaussians.log_likelihood(test_samples)
- expected_likelihoods = np.array([
- [-1.8378770664, -4.3378770664, -10.8378770664], # samples ll On gaussian 0
- [-10.8378770664, -4.3378770664, -1.8378770664], # samples ll On gaussian 1
- ])
- np.testing.assert_almost_equal(log_likelihoods, expected_likelihoods, decimal=10)
-
- # Variances threshold application
- gaussians["variance_thresholds"] = np.array([[1e-5, 1e-5], [1e-3, 1e-3]])
- expected_variances = np.array([[1, 1], [1, 1]])
- np.testing.assert_equal(gaussians["variances"], expected_variances)
- gaussians["variances"] = np.array([[1e-8, 1e-4], [1e-2, 1e-8]])
- expected_variances = np.array([[1e-5, 1e-4], [1e-2, 1e-3]])
- np.testing.assert_equal(gaussians["variances"], expected_variances)
- gaussians["variance_thresholds"] = np.array([[1e-3, 1e-3], [1e-3, 1e-3]])
- expected_variances = np.array([[1e-3, 1e-3], [1e-2, 1e-3]])
- np.testing.assert_equal(gaussians["variances"], expected_variances)
-
- # Initializing with different shapes
- gaussians = Gaussians(means=np.zeros(shape=(2,3)), variances=0, variance_thresholds=1e-4)
- np.testing.assert_equal(gaussians["variances"], np.full(shape=(2,3), fill_value=1e-4))
-
-
-def test_gaussians_attributes_copy():
- """Ensures that the input arrays are correctly copied."""
- means = np.array([[0, 0], [1, 1], [2, 2]], dtype=float)
- variances = np.ones_like(means, dtype=float)
- variance_thresholds = np.full_like(means, 1e-5, dtype=float)
- gaussians = Gaussians(means, variances, variance_thresholds)
-
- means[0,0] = np.nan
- expected_means = np.array([[0, 0], [1, 1], [2, 2]])
- np.testing.assert_equal(gaussians["means"], expected_means)
-
- variances[0,0] = np.nan
- expected_variances = np.ones_like(means, dtype=float)
- np.testing.assert_equal(gaussians["variances"], expected_variances)
-
- variance_thresholds[0,0] = np.nan
- expected_thresholds = np.full_like(means, 1e-5, dtype=float)
- np.testing.assert_equal(gaussians["variance_thresholds"], expected_thresholds)