# Display Options.
log_str = ("Autoencoder-{size} Training\n"
" Running on {num_ipus} IPUs\n"
" Dataset {training_data_file}\n"
" Precision {precision}\n"
" Logging to {logdir}\n"
" Stochastic Rounding {prng}\n"
"Training Graph\n"
" Optimizer {optimizer}\n"
" Batch Size {batch_size}\n"
" Epochs {epochs}\n"
" Base Learning Rates 2^{base_learning_rates}\n")
if opts.loss_scaling:
log_str += " Loss Scaling {loss_scaling}\n"
if opts.weight_decay:
log_str += " Weight Decay {weight_decay}\n"
print(log_str.format(**vars(opts)))
# load data
print("Loading training data")
training_data = AutoencoderData(data_file_name=opts.training_data_file)
print("Users: {}".format(training_data.size))
print("Items: {}".format(training_data.input_size))
opts.input_size = training_data.input_size
train_process_init(opts, training_data)
log_str += " Loss Scaling {loss_scaling}\n"
if opts.weight_decay:
log_str += " Weight Decay {weight_decay}\n"
if not opts.no_validation:
log_str += ("Validation Graph\n"
" Dataset {validation_data_file}\n"
" Batch Size {validation_batch_size}\n")
if not opts.testing_on_checkpoint:
log_str += "Checkpoint Path {checkpoint_path}\n"
opts.learning_rate = (2**opts.base_learning_rate) * opts.batch_size
print(log_str.format(**vars(opts)))
# load data
print("Loading training data")
training_data = AutoencoderData(data_file_name=opts.training_data_file)
print("Users: {}".format(training_data.size))
print("Items: {}".format(training_data.input_size))
print("Loading evaluation data")
valid_data = AutoencoderData(data_file_name=opts.validation_data_file,
training_data=training_data)
print("Users: {}".format(valid_data.size))
print("Items: {}".format(valid_data.input_size))
if training_data.input_size != valid_data.input_size:
raise ValueError(
'Number of items for training data and validation data must be'
' equal. Got {} and {}.'.format(training_data.input_size,
valid_data.input_size))