qbiocode.apps.quvine.analysis.analyze module

Summary

Functions:

cca_correlation
effective_rank
knn_overlap
knn_sets
normalize
plot_singular_values	Plot the log singular-value spectrum, optionally writing it to filename.
procrustes_residual
rsa_corr
spectral_info	Singular-value spectra and effective ranks for a list of embeddings.

Reference

normalize(embedding, eps=1e-08)

procrustes_residual(Z1, Z2)

cca_correlation(Z1, Z2, n_components=10)

rsa_corr(Z1, Z2)

knn_sets(Z, k=10)

knn_overlap(Z1, Z2, k=10)

effective_rank(s)

plot_singular_values(singular_values, label='concatenate', filename=None, show=False)

Plot the log singular-value spectrum, optionally writing it to filename.

Parameters:

• singular_values – Singular values in descending order.

• label – Legend label for the series.

• filename – Where to write the figure; nothing is written when None.

• show – Whether to call plt.show(). Defaults to False because this is library code – under a GUI backend show blocks until a human closes the window, and it is ignored under a non-interactive one.

Returns:

the figure, already closed but still savable.

Return type:

matplotlib.figure.Figure

spectral_info(embeddings, labels, plot_flag=False, outdir='.', show=False)

Singular-value spectra and effective ranks for a list of embeddings.

Parameters:

• embeddings – Embedding matrices to compare.

• labels – One label per embedding, used in the legend and the result keys.

• plot_flag – Whether to draw and write the three spectrum figures.

• outdir – Directory the figures are written to. Previously they went to the process’s current working directory under fixed names, so two runs from the same directory silently overwrote each other’s plots.

• show – Whether to call plt.show(); see plot_singular_values().

Returns:

{label: effective_rank}.

Return type:

dict