torchbearer: A model fitting library for PyTorch
[0.5.5] - 2023-12-01
to_file
to use state information in the file nameepoch
argument in schedulers in newer versions of pytorchImagingCallback.to_pyplot
method, set to false to stop it from calling plt.show
ImagingCallback.to_pyplot
to turn off the axisto_pyplot
was not possiblewith_loader
trial method that allows running of custom batch loaders__call__(state)
to StateKey
so that they can now be used as losseswith_data
trial method that allows passing of train, val and test data in one callstep_on_batch
flag to the early stopping callbackimaging
verbose
argument from the early stopping callbackClassAppearanceModel
callbackImagingCallback
that would sometimes cause make_grid
to throw an errortorchbearer.callbacks.init
with_closure
trial method that allows running of custom closuresbase_closure
function to bases that allows creation of standard training loop closuresImagingCallback
class for callbacks which produce images that can be sent to tensorboard, visdom or a fileCachingImagingCallback
and MakeGrid
callback to make a grid of imagesonly_if
callback decorator a function of self and state rather than just stateNo_grad
and enable_grad
decorators are now also context managerstorchvision
, still required for some functionalityTrial
methods%torchbearer notebook