fit_with_generator: Fit a Model using Generators

In neuroimaginador/dl4ni: Deep Learning Flows for Neuroimaging

Description

this function is used to fit or train a DLmodel by using generator functions for training and validation data.

Usage

fit_with_generator(.model, generator = NULL, steps_per_epoch = NULL,
  train_config = NULL, epochs = 10, starting_epoch = 1,
  validation_data = NULL, validation_steps = NULL,
  validation_config = NULL, keep_best = TRUE, verbose = TRUE,
  metrics_viewer = FALSE, reset_optimizer = FALSE, ...)

Arguments

.model	(DLmodel) The model to train
generator	(function) The generator function for training data, built by create_generator_from_config, Default: NULL
steps_per_epoch	(integer) Number of steps per training epoch, Default: NULL
train_config	(list) The training configuration, the output of create_generator_from_config when inputs are training files, Default: NULL
epochs	(numeric) Maximum number of epochs to train, Default: 10
starting_epoch	(numeric) strating epoch, useful when we want to resume a previous fit, Default: 1
validation_data	(function or matrix) Data for validation. It can be a generator function, built by create_generator_from_config, Default: NULL
validation_steps	(integer) Number of steps of validation per epoch, Default: NULL
validation_config	(list) The testing configuration, the output of create_generator_from_config when inputs are validation files, Default: NULL
keep_best	(logical) Should the training always store the best model up-to-date?, Default: TRUE
verbose	(logical) Provide additional information on training, Default: TRUE
metrics_viewer	(logical) Visualize training loss interactively while fitting?, Default: FALSE
reset_optimizer	(logical) Reset optimizer state after each subepoch?, Default: FALSE
...	extra arguments passed to other functions.

Details

generator, steps_per_epoch, validation_data and validation_steps are completely and automatically determined if one uses the train_config and validation_config parameters, both being the outputs of create_generator_from_config.

Additionally, we can pass this function two arguments: path and prefix, the best model will be stored in the corresponding path with the given prefix (usually something indicative or with a timestamp).

Value

The trained DLmodel.

See Also

callback_lambda

neuroimaginador/dl4ni documentation built on May 3, 2019, 5:47 p.m.