Deep Learning trainers

Base trainer class

class mldas.trainers.base. BaseTrainer ( output_dir = None , gpu = None , distributed = False , rank = 0 , ** kwargs )

Base class for PyTorch trainers. This implements the common training logic, logging of summaries, and checkpoints.

Methods Summary

__init__ ([output_dir, gpu, distributed, rank])
print_model_summary ()	Override as needed
save_summary (summaries)	Save summary information
write_summaries ()
write_checkpoint (checkpoint_id)	Write a checkpoint for the model
train (train_data_loader, epochs[, ...])	Run the model training

Methods Documentation

__init__ ( output_dir = None , gpu = None , distributed = False , rank = 0 , ** kwargs )

print_model_summary ( )

Override as needed

save_summary ( summaries )

Save summary information

write_summaries ( )

write_checkpoint ( checkpoint_id )

Write a checkpoint for the model

train ( train_data_loader , epochs , valid_data_loader = None , test_data_loader = None , ** kwargs )

Run the model training

Classifier training

class mldas.trainers.generic. GenericTrainer ( rank , device , output_dir = None , ** kwargs )

Trainer code for basic classification problems.

Methods Summary

__init__ (rank, device[, output_dir])
build_model (device_ids[, distributed, loss, ...])	Instantiate our model
train_epoch (data_loader[, rounded])	Train for one epoch
evaluate (data_loader, mode[, rounded])	"Evaluate the model
accuracy (batch_output, batch_target[, acc_tol])

Methods Documentation

__init__ ( rank , device , output_dir = None , ** kwargs )

build_model ( device_ids , distributed = False , loss = 'CE' , optimizer = 'SGD' , lr = 0.01 , lr_decay_epoch = [] , lr_decay_ratio = 0.5 , momentum = 0.9 , ** model_args )

Instantiate our model

train_epoch ( data_loader , rounded = False , ** kwargs )

Train for one epoch

evaluate ( data_loader , mode , rounded = False , ** kwargs )

“Evaluate the model

accuracy ( batch_output , batch_target , acc_tol = 20 , ** kwargs )