config

class neuralhydrology.utils.config.Config(yml_path_or_dict: Path | dict, dev_mode: bool = False)

Bases: object

Read run configuration from the specified path or dictionary and parse it into a configuration object.

During parsing, config keys that contain ‘dir’, ‘file’, or ‘path’ will be converted to pathlib.Path instances. Configuration keys ending with ‘_date’ will be parsed to pd.Timestamps. The expected format is DD/MM/YYYY.

Parameters:

yml_path_or_dict (Union[Path, dict]) – Either a path to the config file or a dictionary of configuration values.
dev_mode (bool, optional) – If dev_mode is off, the config creation will fail if there are unrecognized keys in the passed config specification. dev_mode can be activated either through this parameter or by setting dev_mode: True in yml_path_or_dict.

Raises:

ValueError – If the passed configuration specification is neither a Path nor a dict or if dev_mode is off (default) and the config file or dict contain unrecognized keys.

property additional_feature_files: List[Path]

property allow_subsequent_nan_losses: int

as_dict() → dict

Return run configuration as dictionary.

Returns:: The run configuration, as defined in the .yml file.
Return type:: dict

property autoregressive_inputs: List[str] | Dict[str, List[str]]

property base_run_dir: Path

property batch_size: int

property bidirectional_stacked_forecast_lstm: bool

property cache_validation_data: bool

property checkpoint_path: Path

property clip_gradient_norm: float

property clip_targets_to_zero: List[str]

property conceptual_model: str

property continue_from_epoch: int

property custom_normalization: dict

property data_dir: Path

property dataset: str

property device: str

dump_config(folder: Path, filename: str = 'config.yml')

Save the run configuration as a .yml file to disk.

Parameters:

folder (Path) – Folder in which the configuration will be stored.
filename (str, optional) – Name of the file that will be stored. Default: ‘config.yml’.

Raises:

FileExistsError – If the specified folder already contains a file named filename.

property duplicate_features: dict

property dynamic_conceptual_inputs: List[str]

property dynamic_inputs: List[str] | Dict[str, List[str]]

property dynamics_embedding: dict

property epochs: int

property evolving_attributes: List[str]

property experiment_name: str

property finetune_modules: List[str] | Dict[str, str]

property forcings: List[str]

property forecast_hidden_size: int

property forecast_inputs: List[str]

property forecast_network: dict

property forecast_overlap: int

property forecast_seq_length: int

property head: str

property hidden_size: int | Dict[str, int]

property hindcast_hidden_size: int | Dict[str, int]

property hindcast_inputs: List[str]

property hydroatlas_attributes: List[str]

property img_log_dir: Path

property initial_forget_bias: float

property is_continue_training: bool

property is_finetuning: bool

property lagged_features: dict

property learning_rate: Dict[int, float]

property log_interval: int

property log_n_figures: int

property log_tensorboard: bool

property loss: str

property mamba_d_conv: int

property mamba_d_state: int

property mamba_expand: int

property mass_inputs: List[str]

property max_updates_per_epoch: int | None

property mc_dropout: bool

property metrics: List[str] | Dict[str, List[str]]

property model: str

property n_distributions: int

property n_samples: int

property n_taus: int

property negative_sample_handling: str

property negative_sample_max_retries: int

property no_loss_frequencies: list

property num_workers: int

property number_of_basins: int

property ode_method: str

property ode_num_unfolds: int

property ode_random_freq_lower_bound: str

property optimizer: str

property output_activation: str

property output_dropout: float

property per_basin_test_periods_file: Path

property per_basin_train_periods_file: Path

property per_basin_validation_periods_file: Path

property predict_last_n: int | Dict[str, int]

property random_holdout_from_dynamic_features: Dict[str, float]

property rating_curve_file: Path

property regularization: List[str | Tuple[str, float]]

property run_dir: Path

property save_all_output: bool

property save_git_diff: bool

property save_train_data: bool

property save_validation_results: bool

property save_weights_every: int

property seed: int

property seq_length: int | Dict[str, int]

property shared_mtslstm: bool

property state_handoff_network: dict

property static_attributes: List[str]

property statics_embedding: dict

property target_loss_weights: List[float]

property target_noise_std: float

property target_variables: List[str]

property tau_down: float

property tau_up: float

property test_basin_file: Path

property test_end_date: pandas.Timestamp

property test_start_date: pandas.Timestamp

property timestep_counter: bool

property train_basin_file: Path

property train_data_file: Path

property train_dir: Path

property train_end_date: pandas.Timestamp

property train_start_date: pandas.Timestamp

property transfer_mtslstm_states: Dict[str, str]

property transformer_dim_feedforward: int

property transformer_dropout: float

property transformer_nheads: int

property transformer_nlayers: int

property transformer_positional_dropout: float

property transformer_positional_encoding_type: str

property umal_extend_batch: bool

update_config(yml_path_or_dict: Path | dict, dev_mode: bool = False)

Update config arguments.

Useful e.g. in the context of fine-tuning or when continuing to train from a checkpoint to adapt for example the learning rate, train basin files or anything else.

Parameters:

yml_path_or_dict (Union[Path, dict]) – Either a path to the new config file or a dictionary of configuration values. Each argument specified in this file will overwrite the existing config argument.
dev_mode (bool, optional) – If dev_mode is off, the config creation will fail if there are unrecognized keys in the passed config specification. dev_mode can be activated either through this parameter or by setting dev_mode: True in yml_path_or_dict.

Raises:

ValueError – If the passed configuration specification is neither a Path nor a dict, or if dev_mode is off (default) and the config file or dict contain unrecognized keys.

property use_basin_id_encoding: bool

property use_frequencies: List[str]

property validate_every: int

property validate_n_random_basins: int

property validation_basin_file: Path

property validation_end_date: pandas.Timestamp

property validation_start_date: pandas.Timestamp

property verbose: int

Defines level of verbosity.

0: Only log info messages, don’t show progress bars 1: Log info messages and show progress bars

Returns:: Level of verbosity.
Return type:: int

property warmup_period: int

neuralhydrology.utils.config.create_random_name()