Subspace Aligned Classifier¶

class libtlda.suba.SemiSubspaceAlignedClassifier(loss_function='logistic', l2_regularization=None, subspace_dim=1)¶

Class of classifiers based on semi-supervised Subspace Alignment.

Methods contain the alignment itself, classifiers and general utilities.

Examples

>>>> X = np.random.randn(10, 2)
>>>> y = np.vstack((-np.ones((5,)), np.ones((5,))))
>>>> Z = np.random.randn(10, 2)
>>>> clf = SubspaceAlignedClassifier()
>>>> clf.fit(X, y, Z)
>>>> preds = clf.predict(Z)

Methods

`align_classes`(X, Y, Z, u, CX, CZ, V)	Project each class separately.
`find_medioid`(X, Y)	Find point with minimal distance to all other points.
`fit`(X, Y, Z[, u])	Fit/train a classifier on data mapped onto transfer components.
`get_params`()	Get classifier parameters.
`is_pos_def`(A)	Check for positive definiteness.
`predict`(Z[, zscore])	Make predictions on new dataset.
`predict_proba`(Z[, zscore, signed_classes])	Make predictions on new dataset.
`reg_cov`(X)	Regularize covariance matrix until non-singular.
`score`(Z, U[, zscore])	Compute classification error on test set.
`semi_subspace_alignment`(X, Y, Z, u[, …])	Compute subspace and alignment matrix, for each class.

align_classes(X, Y, Z, u, CX, CZ, V)¶

Project each class separately.

Parameters:

X : array: source data set (N samples x D features)
Y : array: source labels (N samples x 1)
Z : array: target data set (M samples x D features)
u : array: target labels (m samples x 2)
CX : array: source principal components (K classes x D features x d subspaces)
CZ : array: target principal components (K classes x D features x d subspaces)
V : array: transformation matrix (K classes x d subspaces x d subspaces)

Returns:

X : array: transformed X (N samples x d features)
Z : array: transformed Z (M samples x d features)

find_medioid(X, Y)¶

Find point with minimal distance to all other points.

Parameters:	X : array data set, with N samples x D features. Y : array labels to select for which samples to compute distances.
Returns:	x : array medioid ix : int index of medioid

fit(X, Y, Z, u=None)¶

Fit/train a classifier on data mapped onto transfer components.

Parameters:	X : array source data (N samples x D features). Y : array source labels (N samples x 1). Z : array target data (M samples x D features). u : array target labels, first column corresponds to index of Z and second column corresponds to actual label (number of labels x 2).
Returns:	None

get_params()¶: Get classifier parameters.

is_pos_def(A)¶

Check for positive definiteness.

A : array: square symmetric matrix.

Returns:	bool whether matrix is positive-definite. Warning! Returns false for arrays containing inf or NaN.

predict(Z, zscore=False)¶

Make predictions on new dataset.

Parameters:	Z : array new data set (M samples x D features) zscore : boolean whether to transform the data using z-scoring (def: false)
Returns:	preds : array label predictions (M samples x 1)

predict_proba(Z, zscore=False, signed_classes=False)¶

Make predictions on new dataset.

Parameters:	Z : array new data set (M samples x D features) zscore : boolean whether to transform the data using z-scoring (def: false)
Returns:	preds : array label predictions (M samples x 1)

reg_cov(X)¶

Regularize covariance matrix until non-singular.

Parameters:	C : array square symmetric covariance matrix.
Returns:	C : array regularized covariance matrix.

score(Z, U, zscore=False)¶

Compute classification error on test set.

Parameters:	Z : array new data set (M samples x D features) zscore : boolean whether to transform the data using z-scoring (def: false)
Returns:	preds : array label predictions (M samples x 1)

semi_subspace_alignment(X, Y, Z, u, subspace_dim=1)¶

Compute subspace and alignment matrix, for each class.

Parameters:	X : array source data set (N samples x D features) Y : array source labels (N samples x 1) Z : array target data set (M samples x D features) u : array target labels, first column is index in Z, second column is label (m samples x 2) subspace_dim : int Dimensionality of subspace to retain (def: 1)
Returns:	V : array transformation matrix (K, D features x D features) CX : array source principal component coefficients CZ : array target principal component coefficients

class libtlda.suba.SubspaceAlignedClassifier(loss_function='logistic', l2_regularization=None, subspace_dim=1)¶

Class of classifiers based on Subspace Alignment.

Methods contain the alignment itself, classifiers and general utilities.

Examples

>>>> X = np.random.randn(10, 2)
>>>> y = np.vstack((-np.ones((5,)), np.ones((5,))))
>>>> Z = np.random.randn(10, 2)
>>>> clf = SubspaceAlignedClassifier()
>>>> clf.fit(X, y, Z)
>>>> preds = clf.predict(Z)

Methods

`align_data`(X, Z, CX, CZ, V)	Align data to components and transform source.
`fit`(X, Y, Z)	Fit/train a classifier on data mapped onto transfer components.
`get_params`()	Get classifier parameters.
`is_pos_def`(A)	Check for positive definiteness.
`predict`(Z[, zscore])	Make predictions on new dataset.
`predict_proba`(Z[, zscore, signed_classes])	Make predictions on new dataset.
`reg_cov`(X)	Regularize covariance matrix until non-singular.
`score`(Z, U[, zscore])	Compute classification error on test set.
`subspace_alignment`(X, Z[, subspace_dim])	Compute subspace and alignment matrix.
`zca_whiten`(X)	Perform ZCA whitening (aka Mahalanobis whitening).

align_data(X, Z, CX, CZ, V)¶

Align data to components and transform source.

Parameters:	X : array source data set (N samples x D features) Z : array target data set (M samples x D features) CX : array source principal components (D features x d subspaces) CZ : array target principal component (D features x d subspaces) V : array transformation matrix (d subspaces x d subspaces)
Returns:	X : array transformed source data (N samples x d subspaces) Z : array projected target data (M samples x d subspaces)

fit(X, Y, Z)¶

Fit/train a classifier on data mapped onto transfer components.

Parameters:	X : array source data (N samples x D features). Y : array source labels (N samples x 1). Z : array target data (M samples x D features).
Returns:	None

get_params()¶: Get classifier parameters.

is_pos_def(A)¶: Check for positive definiteness.

predict(Z, zscore=False)¶

Make predictions on new dataset.

Parameters:	Z : array new data set (M samples x D features) zscore : boolean whether to transform the data using z-scoring (def: false)
Returns:	preds : array label predictions (M samples x 1)

predict_proba(Z, zscore=False, signed_classes=False)¶

Make predictions on new dataset.

Parameters:	Z : array new data set (M samples x D features) zscore : boolean whether to transform the data using z-scoring (def: false)
Returns:	preds : array label predictions (M samples x 1)

reg_cov(X)¶

Regularize covariance matrix until non-singular.

Parameters:	C : array square symmetric covariance matrix.
Returns:	C : array regularized covariance matrix.

score(Z, U, zscore=False)¶

Compute classification error on test set.

Parameters:	Z : array new data set (M samples x D features) zscore : boolean whether to transform the data using z-scoring (def: false)
Returns:	preds : array label predictions (M samples x 1)

subspace_alignment(X, Z, subspace_dim=1)¶

Compute subspace and alignment matrix.

Parameters:	X : array source data set (N samples x D features) Z : array target data set (M samples x D features) subspace_dim : int Dimensionality of subspace to retain (def: 1)
Returns:	V : array transformation matrix (D features x D features) CX : array source principal component coefficients CZ : array target principal component coefficients

zca_whiten(X)¶

Perform ZCA whitening (aka Mahalanobis whitening).

Parameters:	X : array (M samples x D features) data matrix.
Returns:	X : array (M samples x D features) whitened data.