AttributeError: '_TrackedCheckpoint' object has no attribute 'value'

Ray Libraries (Data, Train, Tune, Serve) Ray Tune
1

Hi, I have a quick relevant question.
I am running the code:

result = tune.run(
        tune.with_parameters(train),
        resources_per_trial={"cpu": 12, "gpu": gpus_per_trial},
        config=config,
        num_samples=num_samples,
        search_alg=hyperopt_search,
        scheduler=scheduler,
        progress_reporter=reporter,
        keep_checkpoints_num = 1,
        checkpoint_score_attr = "loss"
        )

    best_trial = result.get_best_trial("accuracy", "max", "last")
    print("Best trial config: {}".format(best_trial.config))
    print("Best trial final validation accuracy: {}".format(best_trial.last_result["accuracy"]))

    best_trained_model = LeNet5().to(device)
    best_checkpoint_dir = best_trial.checkpoint.value
    model_state, optimizer_state = torch.load(os.path.join(best_checkpoint_dir, "checkpoint"))
    best_trained_model.load_state_dict(model_state)

The code was running without throwing an error before but now I am getting this error:

best_checkpoint_dir = best_trial.checkpoint.value
AttributeError: '_TrackedCheckpoint' object has no attribute 'value'

It finds the best config and prints out but cannot get a checkpoint of the best config.
I would be so glad if you help with this issue, thanks =)

2

Hi,
Instead of doing best_trial.checkpoint.value, can you try result.best_checkpoint?

1 Like
3

Hi, thank you so much for such a quick return, I like the Ray Team :slight_smile:
I solved the problem as you also suggested. I am sharing the code so that it can help other people if they have the same question.
best_checkpoint.to_directory(path=“directory”) creates folder named directory and saves the best checkpoint inside of it.
Then we can load that best checkpoint with torch.load(os.path.join(best_checkpoint_dir, “checkpoint”))

result = tune.run(tune.with_parameters(model.update_representation),
                              resources_per_trial={"cpu": 12, "gpu": 1},
                              config=config,
                              num_samples=2,
                              search_alg=hyperopt_search,
                              scheduler=scheduler,
                              keep_checkpoints_num=1,
                              checkpoint_score_attr="loss")

            best_trial = result.get_best_trial("accuracy", "max", "last")
            print("Best trial config: {}".format(best_trial.config))
            print("Best trial final validation loss: {}".format(best_trial.last_result["loss"]))
            print("Best trial final validation accuracy: {}".format(best_trial.last_result["accuracy"]))

            best_checkpoint = result.get_best_checkpoint(trial=best_trial, metric="accuracy", mode="max")
            best_checkpoint_dir = best_checkpoint.to_directory(path="directory")
            model_state, optimizer_state = torch.load(os.path.join(best_checkpoint_dir, "checkpoint"))
            best_trained_model = model._network
            best_trained_model.load_state_dict(model_state)
            model._network.load_state_dict(model_state)
1 Like