def save_results(model, train_loss, dev_loss, train_size, dev_size, results_fname):
results = [['alphabet_size', 'embedding_size', 'hidden_size', 'nlayers',\
'dropout_p', 'train_loss', 'dev_loss',\
'train_size', 'dev_size']]
results += [[model.alphabet_size, model.embedding_size, model.hidden_size,\
model.nlayers, model.dropout_p, train_loss, dev_loss,\
train_size, dev_size]]
util.write_csv(results_fname, results)
def save_results(model, train_loss, dev_loss, test_loss, results_fname):
args = model.get_args()
del args['alphabet']
results = [['name', 'train_loss', 'dev_loss', 'test_loss', 'alphabet_size'] +
list(args.keys())]
results += [[model.name, train_loss, dev_loss, test_loss, model.alphabet_size] +
list(args.values())]
util.write_csv(results_fname, results)
def get_p_values(keep_eos, args, model):
print('\nRunning model: %s - %s' % (model, str(keep_eos)))
(losses, y_values, lengths) = get_results(args.checkpoints_path, keep_eos=keep_eos, models=[model])
results = analyse_languages(losses, y_values, lengths, model_type=model, n_permutations=args.n_permutations)
fname = '%s_%s__%s--%d.tsv' % (args.dataset, model, str(keep_eos), args.n_permutations)
util.write_csv('results/p_values/bin--%s' % fname, results)
def save_results(model, train_results, dev_results, test_results, results_fname):
results = [model.print_param_names() +
['train_loss', 'dev_loss', 'test_loss',
'train_acc', 'dev_acc', 'test_acc',
'train_norm', 'dev_norm', 'test_norm']]
results += [model.print_params() +
[train_results['loss'], dev_results['loss'], test_results['loss'],
train_results['acc'], dev_results['acc'], test_results['acc'],
train_results['norm'], dev_results['norm'], test_results['norm']]]
util.write_csv(results_fname, results)
def save_results(model, train_results, dev_results, test_results,
results_fname):
results = [[
'n_classes', 'embedding_size', 'hidden_size', 'nlayers', 'dropout_p',
'train_loss', 'dev_loss', 'test_loss', 'train_acc', 'dev_acc',
'test_acc'
]]
results += [[
model.n_classes, model.embedding_size, model.hidden_size,
model.nlayers, model.dropout_p, train_results['loss'],
dev_results['loss'], test_results['loss'], train_results['acc'],
dev_results['acc'], test_results['acc']
]]
util.write_csv(results_fname, results)
def save_results(model, atural_code_avg, permuted_natural_avg, two_stage_avg,\
natural_correlation, permuted_correlation, two_stage_correlation,\
alphabet_size, sentences, results_fname, test):
print('Saving to', results_fname)
results = []
file_size = os.path.getsize(results_fname) if os.path.exists(
results_fname) else 0
if file_size == 0:
results = [['model', 'natural_code_avg', 'permuted_natural_code_avg',\
'two_stage_code_avg', 'natural_correlation', 'permuted_correlation',\
'two_stage_correlation', 'alphabet_size', 'sentences', 'test']]
results += [[model, atural_code_avg, permuted_natural_avg, two_stage_avg,\
natural_correlation, permuted_correlation, two_stage_correlation,\
alphabet_size, sentences, test]]
util.write_csv(results_fname, results)
def save_two_stage_training_results(model, args, train_loss, dev_loss, generator_dev_loss,\
training_time, train_size, dev_size):
results_fname = args.adaptor_results_file
print('Saving to', results_fname)
results = []
file_size = os.path.getsize(results_fname) if os.path.exists(
results_fname) else 0
if file_size == 0:
results = [['alphabet_size', 'embedding_size', 'hidden_size', 'nlayers',
'dropout_p', 'alpha', 'beta', 'train_loss', 'dev_loss',\
'generator_dev_losss', 'total_epochs', 'adaptor_iterations',\
'training_time', 'train_size', 'dev_size']]
results += [[model.alphabet_size, model.embedding_size, model.hidden_size, model.nlayers,\
model.dropout_p, args.alpha, args.beta, train_loss, dev_loss,\
generator_dev_loss, args.epochs, args.adaptor_iterations,\
training_time, train_size, dev_size]]
util.write_csv(results_fname, results)
def main():
# pylint: disable=all
args = get_args()
folds = util.get_folds()
trainloader, devloader, _, alphabet = \
get_data_loaders_with_folds('tokens', args.data_file, folds, args.batch_size,\
max_train_tokens=args.max_train_tokens)
print('Train size: %d Dev size %d' %
(len(trainloader.dataset), len(devloader.dataset)))
beta_limit = len(trainloader.dataset) * 2
if args.beta_limit is not None:
beta_limit = args.beta_limit
print('Tuning alpha and beta')
tuning_results = tune_alpha_and_beta(trainloader, devloader, alphabet, args, args.no_iterations, beta_limit)
print('Writing tuning results to', args.results_file)
util.write_csv(args.results_file, tuning_results)