sccross.models.utils

Probability distributions

Functions

autodevice	Get torch computation device automatically based on GPU availability and memory usage
freeze_running_stats	Selectively stops normalization layers from updating running stats
get_default_numpy_dtype	Get numpy dtype matching that of the pytorch default dtype

Classes

EarlyStopping	Early stop model training when loss no longer decreases
LRScheduler	Reduce learning rate on loss plateau
MSE	A “sham” distribution that outputs negative MSE on log_prob
Model	Abstract model class
RMSE	A “sham” distribution that outputs negative RMSE on log_prob
Tensorboard	Training logging via tensorboard
Trainer	Abstract trainer class
TrainingPlugin	Plugin used to extend the training process with certain functions
ZILN	Zero-inflated log-normal distribution with subsetting support
ZIN	Zero-inflated normal distribution with subsetting support
ZINB	Zero-inflated negative binomial distribution