aibedo.test.reload_and_test_model

aibedo.test.reload_and_test_model(run_id: str, checkpoint_path: Optional[str] = None, config: Optional[omegaconf.dictconfig.DictConfig] = None, entity: str = 'salv47', project: str = 'AIBEDO', train_model: bool = False, test_model: bool = True, predict_model: bool = False, override_kwargs: Optional[Sequence[str]] = None)[source]
This function reloads a model from a checkpoint and trains and/or tests it.

  • If train_model is True, it trains the model (i.e. can be used to resume training).

  • If test_model is True, it tests the model (i.e. can be used to test a trained model).

If the model was trained using Wandb logging, reloading it and resuming training or testing will be as easy as:
>>> example_run_id = "22ejv03e"
>>> reload_and_test_model(run_id=example_run_id, train_model=True, test_model=True)
To resume training and train for more epochs (e.g. 100), use the following:
>>> reload_and_test_model(run_id=example_run_id, train_model=True, override_kwargs=['trainer.max_epochs=100'])
Parameters

  • run_id (str) – Wandb run id

  • checkpoint_path (str) – An optional local ckpt path to load the weights from. If None, the best one on wandb will be used.

  • config (DictConfig) – An optional config to load the model and data from. If None, the config is loaded from Wandb.

  • entity (str) – Wandb entity

  • project (str) – Wandb project

  • train_model (bool) – Whether to train the model before (optional) testing. If you want to train the model for more epochs, make sure to override the original number of epochs with e.g. override_kwargs=['trainer.max_epochs=100', ...]. Default: False

  • test_model (bool) – Whether to test the model. Default: True

  • predict_model (bool) – Whether to predict with the model (the predict_dataloader() of the datamodule). Default: False

  • override_kwargs – A list of strings (of the form “key=value”) that will override the given/reloaded config.