Meta experiment functions

Very high-level functions to run experiments with minimal code, directly from terminal.

Warnings

Some warnings are irrelevant for this library

set_warnings

 set_warnings (logging_level)

Set warnings to be ignored for the given logging level or higher.

Load files and set-up tracking

Fist part of experiment: Log into wandb and load config files

source

prep_experiment

 prep_experiment (project_name:str,
                  libraries_to_track:List[str]=['ddopai'],
                  config_train_name:str='config_train',
                  config_agent_name:str='config_agent',
                  config_env_name:str='config_env')

First stpes to always execute when starting an experiment (using wandb for tracking)

source

init_wandb

 init_wandb (project_name:str)

init wandb

	Type	Details
project_name	str

source

track_libraries_and_git

 track_libraries_and_git (libraries_to_track:List[str],
                          tracking:bool=True, tracking_tool='wandb')

Track the versions of the libraries and the git hash of the repository.

	Type	Default	Details
libraries_to_track	List
tracking	bool	True
tracking_tool	str	wandb	Currenty only wandb is supported
Returns	None

source

import_config

 import_config (filename:str, path:str=None)

Import a config file in YAML format

	Type	Default	Details
filename	str		Name of the file, must be a yaml file
path	str	None	Optional path to the file if it is not in the current directory
Returns	Dict

source

transfer_additional_target_to_env

 transfer_additional_target_to_env (config_env:Dict, config_agent:Dict)

Transfer the lag window from the agent configuration to the environment configuration

	Type	Details
config_env	Dict
config_agent	Dict
Returns	None

source

transfer_lag_window_to_env

 transfer_lag_window_to_env (config_env:Dict, config_agent:Dict)

Transfer the lag window from the agent configuration to the environment configuration

	Type	Details
config_env	Dict
config_agent	Dict
Returns	None

Import data

Import data from the ddop package

source

get_ddop_data

 get_ddop_data (config_env:Dict, overwrite:bool=False)

Standard function to load data provided by the ddop package

source

download_data

 download_data (config_env:Dict, overwrite:bool=False)

Download standard dataset from ddop repository using the DatasetLoader class

	Type	Default
config_env	Dict
overwrite	bool	False
Returns	Tuple

source

set_indices

 set_indices (config_env:Dict, X:numpy.ndarray)

Set the indices for the validation and test set

	Type	Details
config_env	Dict
X	ndarray
Returns	Tuple

Set up environment

Some functions to set-up the environment

source

set_up_env

 set_up_env (env_class, raw_data:Tuple, val_index_start:int,
             test_index_start:int, config_env:Dict, postprocessors:List)

Set up the environment

	Type	Details
env_class
raw_data	Tuple
val_index_start	int
test_index_start	int
config_env	Dict
postprocessors	List
Returns	object

Set up training

Some functions to set-up the environment

source

set_up_earlystoppinghandler

 set_up_earlystoppinghandler (config_train:Dict)

Set up the early stopping handler

	Type	Details
config_train	Dict
Returns	object

Testing

Some functions to test the final model.

source

prep_and_run_test

 prep_and_run_test (agent, environment, agent_dir:str=None,
                    save_dataset:bool=True, save_features:bool=False,
                    dataset_dir:str=None, eval_step_info=False,
                    tracking='wandb')

Test the agent in the environment.

Clean-up

Function to clean-up the experiment script

source

clean_up

 clean_up (agent, environment)

Clean up agent and environment to free up GPU memory

Helper functions

Some functions that are needed to run an experiment

source

select_agent

 select_agent (agent_name:str)

Select an agent class from a list of agent names and return the class

	Type	Details
agent_name	str
Returns	type

source

merge_with_namespace

 merge_with_namespace (target_dict, source_dict, target_dict_name)

*Merge source_dict into target_dict, using the keys as namespaces. For example, if target_dict_name is “agent”, the key “agent-epsilon” in source_dict will be merged into target_dict[“epsilon”]. The function is to merge hyperparameters from a config file with the default hyperparameters from the yaml files

Args: target_dict (dict): Target dictionary source_dict (dict): Source dictionary target_dict_name (str): Name of the target dictionary

Returns: dict: Merged dictionary*