Skip to content

Diagnostics

diagnostics

Diagnostics for a fitted qlPCA (thesis section 3.1.4 and 3.1.5; docs/SPEC.md section 5).

vaf

vaf(loadings)

Variance accounted for by each component, sum_j a_js^2.

Equals the s-th eigenvalue of the correlation matrix F'F / n of the transformed variables.

Parameters:

Name Type Description Default
loadings array-like of shape (m, p)

Loadings.

required

Returns:

Type Description
ndarray of shape (p,)

VAF of each component.

vaf_percent

vaf_percent(loadings)

VAF of each component as a percentage of the number of variables (VAF_s * 100 / m).

It refers to the transformed variables; comparing it with PCA percentages needs caution (thesis pp. 60-61).

Parameters:

Name Type Description Default
loadings array-like of shape (m, p)

Loadings.

required

Returns:

Type Description
ndarray of shape (p,)

Percentages.

communalities

communalities(loadings)

Communality of each variable, sum_s a_js^2 (rule of thumb: keep variables >= 0.25).

Parameters:

Name Type Description Default
loadings array-like of shape (m, p)

Loadings.

required

Returns:

Type Description
ndarray of shape (m,)

Communalities.

piecewise_loadings

piecewise_loadings(model, X, min_points=3)

Piecewise loadings of a degree-1 qlPCA (thesis eqs. 3.7-3.8).

For every variable j and every segment between consecutive distinct knots, the vectors are truncated to the observations whose h_j lies in the (closed) segment, and the loading on component s is the correlation of the scores x_s with the original variable on that segment. Because f_j is linear on the segment, corr(x_s, h_j) = sign(slope) corr(x_s, f_j) on the truncated vectors, so the sign of the slope tells whether the original variable is positively or negatively related to the component there.

Parameters:

Name Type Description Default
model QLPCA

A fitted degree-1 model.

required
X array-like of shape (n_samples, n_features)

Data (usually the training data).

required
min_points int

Segments with fewer observations get NaN loadings.

3

Returns:

Type Description
list of dict

One dict per variable with edges (segment limits in the units of the data), slope_sign, loadings (n_segments, p), n (observations per segment) and loadings_via_f (the same computed through f_j and the slope sign).