decode.neuralfitter.train package#

Submodules#

decode.neuralfitter.train.live_engine module#

decode.neuralfitter.train.live_engine.live_engine_setup(param_file, device_overwrite=None, debug=False, no_log=False, num_worker_override=None, log_folder='runs', log_comment=None)[source]#

Sets up the engine to train DECODE. Includes sample simulation and the actual training.

Parameters:

  • param_file (str) – parameter file path

  • device_overwrite (Optional[str]) – overwrite cuda index specified by param file

  • debug (bool) – activate debug mode (i.e. less samples) for fast testing

  • no_log (bool) – disable logging

  • num_worker_override (Optional[int]) – overwrite number of workers for dataloader

  • log_folder (str) – folder for logging (where tensorboard puts its stuff)

  • log_comment (Optional[str]) – comment to the experiment

decode.neuralfitter.train.live_engine.parse_args()[source]#
decode.neuralfitter.train.live_engine.setup_dataloader(param, train_ds, test_ds=None)[source]#

Set’s up dataloader

decode.neuralfitter.train.live_engine.setup_trainer(simulator_train, simulator_test, logger, model_out, ckpt_path, device, param)[source]#

Set model, optimiser, loss and schedulers

decode.neuralfitter.train.random_simulation module#

decode.neuralfitter.train.random_simulation.setup_random_simulation(param)[source]#

Setup the actual simulation

  1. Define PSF function (load the calibration)

  2. Define our struture from which we sample (random prior in 3D) and its photophysics

  3. Define background and noise

  4. Setup simulation and datasets

Module contents#