How to load data structures from logs

Introduction

After a training / testing session has been finished, a Logger will take care of plotting and saving relevant data.

To avoid recomputing and make life easier, the Performance module has a few “loading” functions to load useful data from the logs.

This is usually the next after having found absolute paths of logs using search_paths like described in Searching through logs.

Methods

This method allows the user to load an already trained agent given the path to the log of its training session. This allows to avoid recomputing.

Performance.load_trained_agent

def load_trained_agent(agent: Agent, results_path: str) -> Agent:
    """Given an initialized agent with its environment and an absolute path of a training session,
        loads the trained data structures of this session.

    Args:
        agent (Agent): Initialized agent with its environment
        results_path (str): Absolute path of the log session of the training session

    Returns:
        Agent: Agent with trained data structures of the training session
    """

This method allows to have a quick access to the dictionary of parameters of the logging session.

Performance.load_json_params

def load_json_params(results_path: str) -> Dict[str, Any]:
    """Given an absolute path of a log session, loads the json file of parameters as a dictionary.

    Args:
        results_path (str): Absolute path of a log session

    Returns:
        Dict[str, Any]: Dictionary of the parameters in the json file in results_path
    """

This method allows to load the pandas DataFrame summarizing the entire log session.

Performance.load_summary_df

def load_summary_df(results_path: str) -> pd.DataFrame:
    """Given an absolute path of a log session, loads the summary DataFrame of the session.

    Args:
        results_path (str): Absolute path of a log session

    Returns:
        pd.DataFrame: Summary DataFrame of the training session
    """