def test_ensure_array_dim():
with raises(ValueError):
ensure_ndarray_2D(np.random.rand(10, 5), ensure_num_cols=3)
with raises(ValueError):
ensure_ndarray_2D(np.random.rand(10), ensure_num_cols=3)
with raises(ValueError):
ensure_ndarray_1D(np.random.rand(10, 5))
with raises(ValueError):
ensure_ndarray_1D(np.random.rand(10, 5, 10))
def __init__(self, km_set, weights, name='WeightedAverageKernel'):
"""Constructor."""
super().__init__(km_set, name=name)
if self.km_set.size == len(weights):
self.weights = ensure_ndarray_1D(weights)
else:
raise ValueError('Number of weights ({}) supplied differ from the '
'kernel set size ({})'
''.format(self.km_set.size, len(weights)))
def fit(self, kernel_weights=None):
"""Computes the sum kernel"""
if kernel_weights is None:
kernel_weights = np.ones(self.km_set.size)
else:
kernel_weights = ensure_ndarray_1D(kernel_weights)
if kernel_weights.size != self.km_set.size:
raise ValueError('Incompatible set of kernel_weights given.'
'Must be an array of length exactly {}'
''.format(self.km_set.size))
self.KM = np.zeros((self.num_samples, self.num_samples))
for weight, km in zip(kernel_weights, self.km_set):
self.KM = self.KM + weight * km.full
self._is_fitted = True
def linear_combination(km_set, weights, norm_weights=False):
"""
Weighted linear combinations of a set of given kernel matrices
Parameters
----------
km_set : KernelSet
Collection of compatible kernel matrices
weights : Iterable
Set of weights for the kernel matrices in km_set.
Weights are not checked to sum to 1.0. Use norm_weights=True if needed.
norm_weights : bool
Flag to request normalizing weights to ensure they sum to 1.0
Returns
-------
lin_comb_KM : ndarray
Final result of weighted linear combination of the kernel matrix set
"""
if km_set.size == len(weights):
weights = ensure_ndarray_1D(weights)
else:
raise ValueError('Number of weights ({}) supplied differ '
'from the kernel set size ({})'
''.format(km_set.size, len(weights)))
if norm_weights:
denom = weights.sum()
if np.isclose(denom, 0.0):
raise RuntimeError('sum of weights == 0.0, unable to normalize!')
weights = weights / denom
# Computes the weighted average kernel
# km_set.num_samples is a tuple (N, M) when operating on two samples
# e.g. train x test
KM = np.zeros(km_set.num_samples)
for weight, km in zip(weights, km_set):
KM = KM + weight * km.full
return KM
def linear_combination(km_set, weights):
"""
Weighted linear combinations of a set of given kernel matrices
Parameters
----------
km_set : KernelSet
Collection of compatible kernel matrices
weights : Iterable
Set of weights for the kernel matrices in km_set
Returns
-------
lin_comb_KM : ndarray
Final result of weighted linear combination of the kernel matrix set
"""
if km_set.size == len(weights):
weights = ensure_ndarray_1D(weights)
else:
raise ValueError('Number of weights ({}) supplied differ '
'from the kernel set size ({})'
''.format(km_set.size, len(weights)))
# TODO should we not ensure weights sum to 1.0?
# Computes the weighted average kernel
# km_set.num_samples is a tuple (N, M) when operating on two samples
# e.g. train x test
KM = np.zeros(km_set.num_samples)
for weight, km in zip(weights, km_set):
KM = KM + weight * km.full
return KM