elektronn3.training.train_utils module¶

class elektronn3.training.train_utils.AccumulationArray(right_shape=(), dtype=<class 'numpy.float32'>, n_init=100, data=None, ema_factor=0.95)[source]¶

Bases: object

add_offset(off)[source]¶

append(data)[source]¶

clear()[source]¶

property data¶

property ema¶

max()[source]¶

mean()[source]¶

min()[source]¶

sum()[source]¶

class elektronn3.training.train_utils.HistoryTracker[source]¶

Bases: object

plot(save_path=None, autoscale=True, close=True, loss_smoothing_len=200)[source]¶

register_debug_output_names(names)[source]¶

update_debug_outputs(vals)[source]¶

update_history(vals)[source]¶

update_regression(pred, target)[source]¶

update_timeline(vals)[source]¶

class elektronn3.training.train_utils.Timer[source]¶

Bases: object

property t_passed¶

elektronn3.training.train_utils.create_preview_batch_from_knossos(knossos_preview_config)[source]¶

Return type:: Tensor

elektronn3.training.train_utils.pretty_string_time(t)[source]¶: Custom printing of elapsed time