Trainer: Trainer

Description Usage Arguments Details

Description

The instance of the class should be created by using training_session() function.

Usage

1
Trainer(model, criterion, parameter_learners, progress_writers = NULL)

Arguments

model

- root node of the Function to train

criterion

(list of Function or Variable) - Function with one or two outputs, representing loss and, if given, evaluation metric (in this order)

parameter_learners

(list) – list of learners

progress_writers

(progress writer or list of them) – optionally, list of progress writers to automatically track training progress.

Details

A training session trains a model using the specified trainer and configs. Different aspects of training such as data sources, checkpointing, cross validation, progress printing can be configured using the corresponding config classes.

****** Properties: ******

evaluation_function - The evaluation function that the trainer is using.

loss_function - The loss function that the trainer is using.

model - The model that the trainer is training.

parameter_learners - The parameter learners that the trainer is using.

previous_minibatch_evaluation_average - The average evaluation criterion value per sample for the last minibatch trained

previous_minibatch_loss_average - The average training loss per sample for the last minibatch trained

previous_minibatch_sample_count - The number of samples in the last minibatch trained with

total_number_of_samples_seen - The number of samples seen globally between all workers from the beginning of training.

****** Associated Functions: ******

restore_trainer_from_checkpoint

save_trainer_checkpoint

summarize_training_progress

summarize_test_progress

test_minibatch

train_minibatch


Microsoft/CNTK-R documentation built on May 28, 2019, 1:52 p.m.