chesscog.core.training package¶
Core training functionality.
Submodules¶
chesscog.core.training.create_configs module¶
- chesscog.core.training.create_configs.create_configs(classifier: str, include_centercrop: bool = False)¶
Create the YAML configuration files for all registered models for a classifier.
- Parameters
classifier (str) – the classifier (either “occupancy_classifier” or “piece_classifier”)
include_centercrop (bool, optional) – whether to create two configs per model, one including center crop and one not. Defaults to False.
chesscog.core.training.optimizer module¶
- chesscog.core.training.optimizer.build_optimizer_from_config(optimizer_cfg: recap.config.CfgNode, params: Iterable) torch.optim.optimizer.Optimizer¶
Build an optimizer for neural network training from a configuration.
- Parameters
optimizer_cfg (CN) – the optimizer part of the configuration object
params (typing.Iterable) – the parameters to optimize
- Raises
NotImplementedError – if the desired optimizer is not implemented
- Returns
the built optimizer
- Return type
torch.optim.Optimizer
chesscog.core.training.train module¶
Main implementation of model training.
- chesscog.core.training.train.train(cfg: recap.config.CfgNode, run_dir: pathlib.Path) torch.nn.modules.module.Module¶
Traing a model.
- Parameters
cfg (CN) – the configuration object describing the model, dataset, etc.
run_dir (Path) – where to write tensorboard files, the active YAML file, and the chosen weights
- Returns
the trained model
- Return type
nn.Module
- chesscog.core.training.train.train_model(cfg: recap.config.CfgNode, run_dir: pathlib.Path, model: torch.nn.modules.module.Module, is_inception: bool = False, model_name: Optional[str] = None, eval_on_train: bool = False) torch.nn.modules.module.Module¶
Train a model that has already been loaded.
- Parameters
cfg (CN) – the configuration object describing the model, dataset, etc.
run_dir (Path) – where to write tensorboard files, the active YAML file, and the chosen weights
model (torch.nn.Module) – the loaded model
is_inception (bool, optional) – whether the model is InceptionV3. Defaults to False.
model_name (str, optional) – the name of the model (by default the last component of the run directory). Defaults to None.
eval_on_train (bool, optional) – whether to evaluate on the training set. Defaults to False.
- Returns
the trained model
- Return type
nn.Module