asaplib.reducedim package

Submodules

asaplib.reducedim.dim_reducer module

Methods and functions to perform dimensionality reduction

class asaplib.reducedim.dim_reducer.Dimension_Reducers(dreduce_spec_dict={})

Bases: object

add(dreduce_spec, tag)

adding the specifications of a new dimensionality reducer :param dreduce_spec: :type dreduce_spec: a dictionary that specify which dimensionality reducer to use

bind()

binds the objects that actually performs dimension reduction these objects need to have .fit(desc)/.transform(desc)/.fit_transform(desc) methods to perform reduction on a design matrix desc

fit(X)

compute the global descriptor vector for a frame from atomic contributions :param X: Input points. :type X: array-like, shape=[n_samples,n_dim_high]

fit_transform(X)

compute the global descriptor vector for a frame from atomic contributions :param X: Input points. :type X: array-like, shape=[n_samples,n_dim_high]

Returns

x

Return type

array-like, shape=[n_samples,n_dim_low]

pack()

transform(X)

compute the global descriptor vector for a frame from atomic contributions :param X: Input points. :type X: array-like, shape=[n_samples,n_dim_high]

Returns

x

Return type

array-like, shape=[n_samples,n_dim_low]

asaplib.reducedim.ml_kpca module

Tools for doing kernel PCA on environmental similarity e.g. kNN = np.genfromtxt(prefix+”.k”,skip_header=1) proj = KernelPCA(kpca_d).fit_transform(kNN)

KernelPCA with precomputed kernel for now only!!!!!!!!!!!!

class asaplib.reducedim.ml_kpca.KernelPCA(n_components=2)

Bases: object

static center_square(kernel)

Centering of a kernel matrix, with additional centering info

Parameters

kernel (array-like, (shape=MxM)) – Kernel matrix

Returns

• colmean (ndarray, (shape=M)) – mean of columns from kernel matrix

• mean (float) – mean of the entire kernel matrix

• centered_kernel (ndarray, (shape=MxM)) – kernel centered in feature space

fit(kmat)

Fit kernel PCA on the precomputed kernel matrix

Notes

• Keeps the kernel matrix intact

• can be done only once on a given object, need to reinitialise if trying to run again

Parameters

kmat (numpy.ndarray, shape=(M,M)) – kerenl matrix between the observations needs to be square and real, symmetric

fit_transform(kmat, copy=True)

Fit kernel PCA on the precomputed kernel matrix & project to lower dimensions

Notes

• Keeps the kernel matrix intact

• can be done only once on a given object, need to reinitialise if trying to run again

Parameters

• kmat (numpy.ndarray, shape=(M,M)) – kerenl matrix between the observations needs to be square and real, symmetric

• copy (bool, optional, default=True) – copy the kernel matrix or overwrite it, passed to self.transform() nb. the kernel matrix will be left centered if this is False

fit_vectors(vecs)

Fit Kernel PCA from vectors in the large dimension

for future, not implemented yet

transform(ktest, iscentered=False, copy=True)

Transforms to the lower dimensions

Parameters

• ktest (array_like, shape=(L,M)) – kernel matrix of the test vectors with the training vectors

• iscentered (bool, optional, default=False) – if the kernel is centered already, mainly used for the fit_transform function

• copy (bool, optional, default=True) – copy the kernel matrix or overwrite it nb. the kernel matrix will be left centered if this is False

Returns

projected_vectors – projections of the given kernel wrt. the training kernel

Return type

numpy.array, shape=(L,n_components)

transform_vectors(vecs)

Fit Kernel PCA from vectors in the large dimension & project to lower dimension

for future, not implemented yet

asaplib.reducedim.ml_pca module

tools for doing PCA e.g. desc = np.genfromtxt(prefix+”.desc”) proj = PCA(pca_d).fit_transform(desc)

class asaplib.reducedim.ml_pca.PCA(n_components=2, scalecenter=True)

Bases: object

centering(desc)

fit(desc)

Fit PCA on the precomputed descriptor matrix

Parameters

• desc (array-like, shape=[n_descriptors, n_samples]) –

• matrix (design) –

fit_transform(desc)

Fit PCA on the design matrix & project to lower dimensions

Parameters

desc_test (array_like, shape=[n_descriptors, n_samples]) – design matrix of the new samples

Returns

projected_vectors – projections of the design matrix on low dimension

Return type

numpy.array, shape=(L,n_components)

scalecenter_matrix(desc)

Scaling and Centering of a design matrix, with additional centering info

Parameters

desc (array-like, shape=[n_descriptors, n_samples]) – design matrix

transform(desc_test)

Transforms to the lower dimensions

Parameters

desc_test (array_like, shape=[n_descriptors, n_samples]) – design matrix of the new samples

Returns

projected_vectors – projections of the design matrix on low dimension

Return type

numpy.array, shape=(L,n_components)

asaplib.reducedim.sparse_kpca module

Tools for performing dimensionality reduction of design matrices using sparse KPCA

class asaplib.reducedim.sparse_kpca.SPARSE_KPCA(n_components=2, kernel={}, sparse_mode='fps', n_sparse=None)

Bases: object

fit(desc)

Fit KPCA on the precomputed descriptor matrix

Parameters

• desc (array-like, shape=[n_descriptors, n_samples]) –

• matrix (design) –

fit_transform(desc)

Fit PCA on the design matrix & project to lower dimensions

Parameters

desc (array_like, shape=[n_descriptors, n_samples]) – design matrix of the new samples

Returns

projected_vectors – projections of the design matrix on low dimension

Return type

numpy.array, shape=(L,n_components)

transform(desc_test)

Transforms to the lower dimensions

Parameters

desc_test (array_like, shape=[n_descriptors, n_samples]) – design matrix of the new samples

Returns

projected_vectors – projections of the design matrix on low dimension

Return type

numpy.array, shape=(L,n_components)

Module contents