Figure 1
Graph of ugtm v2.0 modules: (1) ugtm_classes: classes for generative topographic mapping (GTM) models, (2) ugtm_core: kernel GTM (kGTM) and GTM core functions, (3) ugtm_gtm: expectation-maximization algorithm for GTM, (4) ugtm_kgtm: expectation-maximization algorithm for kGTM, (5) ugtm_landscape: functions for colouring maps, (6) ugtm_predictions: GTM-based prediction algorithms, (7) ugtm_sklearn: sklearn-compatible eGTM transformer, eGTC classifier, and eGTR regressor, (8) ugtm_preprocess: preprocessing functions for data scaling and PCA preprocessing, using sklearn, (9) ugtm_plot: plotting functions for GTM maps, using matplotlib and mpld3, (10) ugtm_crossvalidate: cross-validation workflows.
Figure 2
Generative topographic mapping (GTM) representations of the S curve dataset (downloaded from sklearn): mean positions, modes, and landscape for continuous labels. The code to reproduce this plot is accessible online (https://ugtm.readthedocs.io/en/latest/visualization_examples.html). The GTM projection can be compared to t-distributed stochastic neighbor embedding (t-SNE), multidimensional scaling (MDS) or locally linear embedding (LLE). The axes x1 and x2 are latent axes found by the corresponding algorithm.
Figure 3
GTM representations of the hand-written digits dataset (digits 0 to 5, from the UCI database): mean positions, modes, and class map for discrete labels. The code to reproduce this plot is accessible online (https://ugtm.readthedocs.io/en/latest/visualization_examples.html). The GTM projection can be compared to t-distributed stochastic neighbor embedding (t-SNE), multidimensional scaling (MDS) or locally linear embedding (LLE). The axes x1 and x2 are latent axes found by the corresponding algorithm.