Simple Logger

The Simple Logger is the basic logger included in ICCP. This logger works in conjunction with the Simple Environment.

Class methods and definitions

Initialization

As said in the Logger class description, a logger is instantiated for a single training/testing session and is responsible for logging all the important data from the session.

When initializing the Simple Logger, given this signature :

def __init__(self,
 logging_path: str,
 agent_name: str, 
 num_episodes: int,
 num_iterations: int):

It will create the following folders inside the of the logging path:

┣📂{agent_name} ┃┣ 📂{current_time} ┃ ┃ ┣ 📂results_{current_time} ┃ ┃ ┃ ┣ 📂experiments_csv (where all the observations and actions are saved) ┃ ┃ ┃ ┣ 📂model_weights (neural networks weights) ┃ ┃ ┃ ┗ 📂plots ┃ ┃ ┃ ┃ ┣ 📂pmv_categories (plots of the time spent by the agent in different pmv categories) ┃ ┃ ┃ ┃ ┗ 📂summary (summarizing plots of all the observations and actions throughout time)

### plot_and_logging

This function is usually used at the end of train() or testand called by the agent itself.

     def plot_and_logging(
        self,
        summary_df: pd.DataFrame,
        agent: Agent,
        episode_num: int,
        is_summary: bool,
        opts: Dict[str, Dict[str, Any]],
    ):
        """Main method of the logger that takes care of saving/logging all the data and plotting
        relevant summaries of the data.

        Args:
            summary_df (pd.DataFrame): Should contain all the relevant data needed to log and plot.
            agent (Agent): The agent that was trained/tested/assessed
            episode_num (int): The current episode number of the session (the logger can be called at the end of every episode )
            is_summary (bool): Boolean flag to make sure the titles make sense.
            opts (Dict[str, Dict[str, Any]]): Dictionary defining the plotting parameters.

        """

Nonetheless, we will explain how to use it:

1. Decide what will the summary_df DataFrame contain. It will probably contain some parameters of Simple Environment such as “Tair” or “PMV”, and the action taken at each step “Tset”. Lastly, you may wish to keep track of agent’s specific parameters such as “Loss”.

2. is_summary should be equal to True, if you’re calling this function at the end of your last training / testing episode and would like plots that summarize the entire training / testing session.

3. opts is the most important parameter. It defines what columns of summary_df will be plotted and further options for each column to obtain a neat plot.

Here’s an example of an opts dictionary:

opts = {
            "Tair": {"secondary_y": None,
             "range": [10, 24],
              "unit": "(°C)",
                    },

            "Tset": {
                "secondary_y": "moving_average",
                "range": [14, 22],
                "unit": "(°C)",
                    }
        }

The plotting options have the following format:

"{column_name}" : {
    "secondary_y": None or "cumulative" or "moving_average"
    "range": None or [low,high]
    "unit": any string describing the unit .e.g "(kJ)"
    }

For column_name:

The column name should be identical to the one used in summary_df.

For "secondary_y":

This defines whether there is an auxiliary plot to be drawn on the same plot as the column.

• None -> No auxiliary plot
• “cumulative” -> auxiliary plot is the cumulative value over time (e.g. cumulative heating)
• “moving_average” -> auxiliary plot is a moving average over a 24 timesteps window

For "range":

This defines the range of values on the y-axis that will be plotted.

• None -> no specified range, the range will be from the minimum to the maximum value
• [low,high] -> The range will go from low to high on the y-axis