def log_validation_results(trainer):
evaluate_validate_timer.resume()
evaluator_validate.run(val_loader)
evaluate_validate_timer.pause()
evaluate_validate_timer.step()
metrics = evaluator_validate.state.metrics
timestamp = get_readable_time()
print(
timestamp +
" Validation set Results - Epoch: {} Avg mae: {:.2f} Avg mse: {:.2f} Avg loss: {:.2f}"
.format(trainer.state.epoch, metrics['mae'], metrics['mse'], 0))
experiment.log_metric("valid_mae", metrics['mae'])
experiment.log_metric("valid_mse", metrics['mse'])
# timer
experiment.log_metric("evaluate_valid_timer",
evaluate_validate_timer.value())
print("evaluate_valid_timer ", evaluate_validate_timer.value())
# check if that validate is best
flag_mae = best_mae.checkAndRecord(metrics['mae'], metrics['mse'])
flag_mse = best_mse.checkAndRecord(metrics['mae'], metrics['mse'])
if flag_mae or flag_mse:
experiment.log_metric("valid_best_mae", metrics['mae'])
experiment.log_metric("valid_best_mse", metrics['mse'])
experiment.log_metric("valid_best_epoch", trainer.state.epoch)
print("BEST VAL, evaluating on test set")
evaluate_test_timer.resume()
evaluator_test.run(test_loader)
evaluate_test_timer.pause()
evaluate_test_timer.step()
test_metrics = evaluator_test.state.metrics
timestamp = get_readable_time()
print(
timestamp +
" Test set Results - Epoch: {} Avg mae: {:.2f} Avg mse: {:.2f} Avg loss: {:.2f}"
.format(trainer.state.epoch, test_metrics['mae'],
test_metrics['mse'], 0))
experiment.log_metric("test_mae", test_metrics['mae'])
experiment.log_metric("test_mse", test_metrics['mse'])
experiment.log_metric("evaluate_test_timer",
evaluate_test_timer.value())
print("evaluate_test_timer ", evaluate_test_timer.value())
def log_validation_results(trainer):
evaluator.run(val_loader)
metrics = evaluator.state.metrics
timestamp = get_readable_time()
print(
timestamp +
" Validation set Results - Epoch: {} Avg mae: {:.2f} Avg mse: {:.2f} Avg loss: {:.2f}"
.format(trainer.state.epoch, metrics['mae'], metrics['mse'],
metrics['nll']))
def log_validation_results(trainer):
evaluator.run(test_loader)
metrics = evaluator.state.metrics
timestamp = get_readable_time()
print(timestamp + " Validation set Results - Epoch: {} Avg mae: {:.2f} Avg mse: {:.2f} Avg loss: {:.2f}"
.format(trainer.state.epoch, metrics['mae'], metrics['mse'], metrics['loss']))
experiment.log_metric("valid_mae", metrics['mae'])
experiment.log_metric("valid_mse", metrics['mse'])
experiment.log_metric("valid_loss", metrics['loss'])
def log_training_results(trainer):
evaluator.run(train_loader)
metrics = evaluator.state.metrics
timestamp = get_readable_time()
print(timestamp + " Training set Results - Epoch: {} Avg mae: {:.2f} Avg mse: {:.2f} Avg loss: {:.2f}"
.format(trainer.state.epoch, metrics['mae'], metrics['mse'], metrics['loss']))
experiment.log_metric("epoch", trainer.state.epoch)
experiment.log_metric("train_mae", metrics['mae'])
experiment.log_metric("train_mse", metrics['mse'])
experiment.log_metric("train_loss", metrics['loss'])
experiment.log_metric("lr", get_lr(optimizer))
def log_training_results(trainer):
experiment.log_metric("epoch", trainer.state.epoch)
if not args.skip_train_eval:
evaluator_train.run(train_loader_eval)
metrics = evaluator_train.state.metrics
timestamp = get_readable_time()
print(
timestamp +
" Training set Results - Epoch: {} Avg mae: {:.2f} Avg mse: {:.2f} Avg loss: {:.2f}"
.format(trainer.state.epoch, metrics['mae'], metrics['mse'], 0)
)
# experiment.log_metric("epoch", trainer.state.epoch)
experiment.log_metric("train_mae", metrics['mae'])
experiment.log_metric("train_mse", metrics['mse'])
experiment.log_metric("lr", get_lr(optimizer))
print("batch_timer ", batch_timer.value())
print("train_timer ", train_timer.value())
experiment.log_metric("batch_timer", batch_timer.value())
experiment.log_metric("train_timer", train_timer.value())
def log_training_loss(trainer):
timestamp = get_readable_time()
print(timestamp + " Epoch[{}] Loss: {:.2f}".format(trainer.state.epoch, trainer.state.output))
test_metrics['mse'], 0))
experiment.log_metric("test_mae", test_metrics['mae'])
experiment.log_metric("test_mse", test_metrics['mse'])
experiment.log_metric("evaluate_test_timer",
evaluate_test_timer.value())
print("evaluate_test_timer ", evaluate_test_timer.value())
# experiment.log_metric("test_loss", test_metrics['loss'])
def checkpoint_valid_mae_score_function(engine):
score = engine.state.metrics['mae']
return -score
if args.eval_only:
print("evaluation only, no training")
timestamp = get_readable_time()
# if flag_mae or flag_mse:
# experiment.log_metric("valid_best_mae", metrics['mae'])
# experiment.log_metric("valid_best_mse", metrics['mse'])
# print("BEST VAL, evaluating on test set")
evaluate_test_timer.resume()
evaluator_test.run(test_loader)
evaluate_test_timer.pause()
evaluate_test_timer.step()
test_metrics = evaluator_test.state.metrics
timestamp = get_readable_time()
if args.eval_density:
print(
timestamp +
