commands.train_vae

Train a VAE for heterogeneous reconstruction with known poses.

Example usage

$ cryodrgn train_vae projections.mrcs -o outs/002_trainvae –lr 0.0001 –zdim 10 –poses angles.pkl –ctf test_ctf.pkl -n 50

# Restart after already running the same command with some epochs completed $ cryodrgn train_vae projections.mrcs -o outs/002_trainvae –lr 0.0001 –zdim 10 –poses angles.pkl –ctf test_ctf.pkl –load latest -n 100

# cryoDRGN-ET tilt series reconstruction $ cryodrgn train_vae particles_from_M.star –datadir particleseries -o your-outdir –ctf ctf.pkl –poses pose.pkl –encode-mode tilt –dose-per-tilt 2.93 –zdim 8 –num-epochs 50 –beta .025

Functions

add_args(parser)
eval_z(model, lattice, data, batch_size, device)
get_latest(args)
loss_function(z_mu, z_logvar, y, ntilts, ...)
main(args)
preprocess_input(y, lattice, trans)
run_batch(model, lattice, y, rot, ntilts[, ...])
save_checkpoint(model, optim, epoch, z_mu, ...)	Save model weights, latent encoding z, and decoder volumes
save_config(args, dataset, lattice, model, ...)
train_batch(model, lattice, y, ntilts, rot, ...)