config Module

The config module defines utilities for managing configurations, including loading and validation.

class ucs.utils.config.Config(dataset=<factory>, directories=<factory>, training=<factory>, callbacks=<factory>)

Parameters:

• dataset (DatasetConfig)

• directories (DirectoriesConfig)

• training (TrainingConfig)

• callbacks (CallbacksConfig)

callbacks: CallbacksConfig

dataset: DatasetConfig

directories: DirectoriesConfig

classmethod load_config(config_path=None, **overrides)

Load YAML configuration file and apply overrides.

Parameters:

• config_path (Optional[str]) – Path to the configuration YAML file.

• **overrides – Dictionary of command-line overrides.

Returns:

An instance of the Config class with loaded values.

Return type:

Config

training: TrainingConfig

class ucs.utils.config.TrainingConfig(model_name='b0', max_epochs=50, learning_rate=2e-05, weight_decay=0.001, ignore_index=0, weighting_strategy='raw', alpha=0.7, id2label=<factory>)

Parameters:

• model_name (str)

• max_epochs (int)

• learning_rate (float)

• weight_decay (float)

• ignore_index (int | None)

• weighting_strategy (str)

• alpha (float)

• id2label (Dict[int, str])

alpha: float = 0.7

id2label: Dict[int, str]

ignore_index: int | None = 0

learning_rate: float = 2e-05

max_epochs: int = 50

model_name: str = 'b0'

weight_decay: float = 0.001

weighting_strategy: str = 'raw'

class ucs.utils.config.DatasetConfig(dataset_path='erikpinhasov/landcover_dataset', batch_size=16, num_workers=8, do_reduce_labels=False, pin_memory=True, model_name=None)

Parameters:

• dataset_path (str)

• batch_size (int)

• num_workers (int)

• do_reduce_labels (bool)

• pin_memory (bool)

• model_name (str | None)

batch_size: int = 16

dataset_path: str = 'erikpinhasov/landcover_dataset'

do_reduce_labels: bool = False

model_name: str | None = None

num_workers: int = 8

pin_memory: bool = True

class ucs.utils.config.DirectoriesConfig(models='models', pretrained='models/pretrained_models', logs='models/logs', checkpoints='models/logs/checkpoints', results='results')

Parameters:

• models (str)

• pretrained (str)

• logs (str)

• checkpoints (str)

• results (str)

checkpoints: str = 'models/logs/checkpoints'

logs: str = 'models/logs'

models: str = 'models'

pretrained: str = 'models/pretrained_models'

results: str = 'results'

class ucs.utils.config.CallbacksConfig(early_stop_patience=5, early_stop_monitor='val_loss', early_stop_mode='min', save_model_monitor='val_mean_iou', save_model_mode='max')

Parameters:

• early_stop_patience (int)

• early_stop_monitor (str)

• early_stop_mode (str)

• save_model_monitor (str)

• save_model_mode (str)

early_stop_mode: str = 'min'

early_stop_monitor: str = 'val_loss'

early_stop_patience: int = 5

save_model_mode: str = 'max'

save_model_monitor: str = 'val_mean_iou'