lm_models= []
lm_model_params = []
## loading the language models
for i,(lm_type,path,order) in enumerate(\
zip(\
arguments['--lm_predictors'].split(','),\
arguments["--lm_path"].split(','),\
[int(o) for o in arguments["--lm_order"].split(',')]\
)):
print '...Loading lm model {}'.format(i)
lm_model_folder = check_path(path, 'LM_MODEL_FOLDER_{}'.format(i), is_data_path=False)
if lm_type=="srilm_char":
lm_model = SRILM_char_lm_loader(path, order)
elif lm_type=="srilm_morph":
lm_model = SRILM_morpheme_lm_loader(path,order)
else:
print "WARNING -- Could not load language model. Unknown type",lm_type,". Use 'srilm_char' or 'srilm_morph'"
lm_models.append(lm_model)
lm_number = len(lm_models)
output_file_path = os.path.join(model_folder,arguments['--pred_path'])
# save best dev model parameters and predictions
print 'Evaluating on test..'
t = time.clock()
accuracy, test_results = evaluate_syncbeam(test_data.iter(indices=[0]), ed_models, lm_models, lm_weights, int(arguments['--beam']))
print 'Time: {}'.format(time.clock()-t)
print 'accuracy: {}'.format(accuracy)
write_pred_file(output_file_path, test_results)
write_eval_file(output_file_path, accuracy, test_path)
if patience == train_hyperparams['PATIENCE']:
print 'out of patience after {} epochs'.format(epoch)
train_progress_bar.finish()
break
# finished epoch
train_progress_bar.update(epoch)
print 'finished training.'
ti = SoftAttention(pc, model_hyperparams, best_model_path)
dev_accuracy, dev_results = ti.evaluate(dev_data.iter(), int(arguments['--beam']))
print 'Best dev accuracy: {}'.format(dev_accuracy)
write_param_file(output_file_path, dict(model_hyperparams.items()+train_hyperparams.items()))
write_pred_file(output_file_path, dev_results)
write_eval_file(output_file_path, best_dev_accuracy, dev_path)
elif arguments['test']:
print '=========EVALUATION ONLY:========='
# requires test path, model path of pretrained path and results path where to write the results to
assert arguments['--test_path']!=None
print 'Loading data...'
test_path = check_path(arguments['--test_path'], '--test_path')
data_set = SoftDataSet
input_format = [int(col) for col in arguments['--input_format'].split(',')]
test_data = data_set.from_file(test_path,input_format, arguments['--lowercase'])
print 'Test data has {} examples'.format(test_data.length)
print 'Checking if any special symbols in data...'
data = set(test_data.inputs + test_data.outputs)
).format(epoch, avg_train_loss, avg_dev_loss, train_perplexity, dev_perplexity, best_dev_perplexity)
log_to_file(log_file_name, epoch, train_perplexity, dev_perplexity)
# if patience == max_patience:
# print 'out of patience after {} epochs'.format(epoch)
# train_progress_bar.finish()
# break
# finished epoch
train_progress_bar.update(epoch)
print 'finished training.'
# save best dev model parameters
write_param_file(output_file_path, dict(model_hyperparams.items()+train_hyperparams.items()))
write_eval_file(output_file_path, best_dev_perplexity, dev_path, 'Perplexity')
elif arguments['test']:
print '=========EVALUATION ONLY:========='
# requires test path, model path of pretrained path and results path where to write the results to
assert arguments['--test_path']!=None
print 'Loading data...'
over_segs = arguments['--segments']
test_path = check_path(arguments['--test_path'], '--test_path')
if arguments['--segformat']:
col_format=3
elif arguments['--dictformat']:
col_format=1
else:
col_format=2