def test_extract_predict_runs(in_file, model_folder, predictions_file):
df_transactions = pd.read_csv(in_file)
df_features = extract_features(df_transactions, model_folder=model_folder)
df_preds = predict(df_features, model_folder=model_folder)
df_targets = extract_targets(df_transactions, model_folder=model_folder)
df_preds_w_targets = df_preds.join(df_targets)
df_preds_w_targets.to_csv(predictions_file)
def main():
args = parse_arguments()
if args.task not in ['haystack', 'form', 'issue', 'target', 'multi']:
print("User error: --task must be either " +\
"haystack, form, issue, target or multi.")
sys.exit(0)
# Read the article as string into variable
with open(args.article, 'r') as file:
text = file.read().replace('\n', '')
model, tokenizer = load_model(args)
# Tokenize article as input to model
text_tokenized = torch.tensor([tokenizer.encode(text)])
results = predict(model, text_tokenized, args)
print_results(results, args)
if args.output_path != '':
store_results(results, args)
for name, param in net.collect_params().items():
try:
sw.add_histogram(tag = name + "_grad", values = param.grad(), global_step = global_step, bins = 1000)
except:
print(name)
print(param.grad())
# compute validation loss
compute_val_loss(net, val_loader, loss_function, sw, epoch)
# evaluate the model on testing set
evaluate(net, test_loader, true_value, num_of_vertices, sw, epoch)
params_filename = os.path.join(params_path, '%s_epoch_%s.params'%(model_name, epoch))
net.save_parameters(params_filename)
print('save parameters to file: %s'%(params_filename))
# close SummaryWriter
sw.close()
if 'prediction_filename' in training_config:
prediction_path = training_config['prediction_filename']
prediction = predict(net, test_loader)
np.savez_compressed(
os.path.normpath(prediction_path),
prediction = prediction,
ground_truth = all_data['test']['target']
)
evaluate(net, test_loader, true_value, supports, device, epoch)
params_filename = os.path.join(
params_path, '%s_epoch_%s_%s.params' %
(model_name, epoch, str(round(valid_loss, 2))))
torch.save(net.state_dict(), params_filename)
print('save parameters to file: %s' % (params_filename))
print("Training finished")
print("Training time/epoch: %.2f secs/epoch" % np.mean(train_time))
bestid = np.argmin(his_loss)
print("The valid loss on best model is epoch%s_%s" %
(str(bestid + 1), str(round(his_loss[bestid], 4))))
best_params_filename = os.path.join(
params_path, '%s_epoch_%s_%s.params' %
(model_name, str(bestid + 1), str(round(his_loss[bestid], 2))))
net.load_state_dict(torch.load(best_params_filename))
start_time_test = time()
prediction, spatial_at, parameter_adj = predict(net, test_loader, supports,
device)
end_time_test = time()
evaluate(net, test_loader, true_value, supports, device, epoch)
test_time = np.mean(end_time_test - start_time_test)
print("Test time: %.2f" % test_time)
np.savez_compressed(os.path.normpath(prediction_path),
prediction=prediction,
ground_truth=all_data['test']['target'])
params_filename = os.path.join(
params_path, '%s_epoch_%s_%s.params' %
(model_name, epoch, str(round(valid_loss, 2))))
torch.save(net.state_dict(), params_filename)
print('save parameters to file: %s' % (params_filename))
print("Training finished")
print("Training time/epoch: %.2f secs/epoch" % np.mean(train_time))
bestid = np.argmin(his_loss)
print("The valid loss on best model is epoch%s_%s" %
(str(bestid + 1), str(round(his_loss[bestid], 4))))
best_params_filename = os.path.join(
params_path, '%s_epoch_%s_%s.params' %
(model_name, str(bestid + 1), str(round(his_loss[bestid], 2))))
net.load_state_dict(torch.load(best_params_filename))
start_time_test = time()
prediction, spatial_at, temporal_at = predict(net, test_loader, supports,
device)
end_time_test = time()
evaluate(net, test_loader, true_value, supports, device, epoch)
test_time = np.mean(end_time_test - start_time_test)
print("Test time: %.2f" % test_time)
np.savez_compressed(os.path.normpath(prediction_path),
prediction=prediction,
spatial_at=spatial_at,
temporal_at=temporal_at,
ground_truth=all_data['test']['target'])
