tomodrgn.convergence

Functions to aid estimation of model training convergence.

Functions

calc_ccs_alltoall_intraepoch	For each epoch in epochs, calculates the (auto-masked) correlation coefficient between each pair of volumes generated at that epoch.
calc_ccs_alltogroundtruth	For each epoch in epochs, calculates the (auto-masked) correlation coefficient between each tomoDRGN volume and each ground_truth volume.
calc_ccs_pairwise_epochs	For each successive pair of epochs in epochs, calculates the (auto-masked) correlation coefficient between the same particle's tomoDRGN-evaluated volume at those epochs.
calc_fscs_pairwise_epochs	For each successive pair of epochs in epochs, calculates the (auto-masked) FSC between the same particle's tomoDRGN-evaluated volume at those epochs.
calc_kld_two_gaussians	Compute the KLD between two gaussians with diagomal covariance.
calc_test_train_pair_volumes_cc_complement	Calculate the scale of heterogeneity among all volumes generated from test latent embeddings of the same particles in the same epoch
calc_test_train_pair_volumes_fscs	Calculate the FSC between volumes generated from test and train latent embeddings of the same particles in the same epoch.
encoder_latent_shifts	Calculates and plots various metrics characterizing the per-particle latent vectors between successive epochs ranging between 0 and final_epoch.
follow_candidate_particles	Plot how the labeled set of particles migrates within latent space at selected epochs over training.
fsc_referencevol_to_manyvols	Calculate the Fourier Shell Correlation (FSC) between one reference volume and many query volumes.
generate_test_train_pair_volumes	Select random particle indices and generate corresponding volumes using train-split images' latent embeddings and test-split latent embeddings.
plot_latent_pca	Calculates, saves, and plots PC1 vs PC2 for all particles' latent embeddings at selected epochs
plot_latent_umap	Calculates, plots, and saves UMAP embeddings of subset of particles' selected epochs' latent encodings
plot_loss	Plot the total loss, reconstruction loss, and KLD divergence per epoch.
sketch_via_umap_local_maxima	Sketch one epoch's latent space via local maxima finding to find dense neighborhoods of particles with similar embeddings as "well-supported" by the data