autora.theorist.bms.utils

present_results(model, model_len, desc_len)

Prints out the best equation, its description length, along with a plot of how this has progressed over the course of the search tasks

Parameters:

Name	Type	Description	Default
model	Tree	The equation which best describes the data	required
model_len	float	The equation loss (defined as description length)	required
desc_len	List[float]	Record of equation loss over time	required

Returns: Nothing

Source code in temp_dir/bms/src/autora/theorist/bms/utils.py

def present_results(model: Tree, model_len: float, desc_len: List[float]) -> None:
    """
    Prints out the best equation, its description length,
    along with a plot of how this has progressed over the course of the search tasks

    Args:
        model: The equation which best describes the data
        model_len: The equation loss (defined as description length)
        desc_len: Record of equation loss over time

    Returns: Nothing

    """
    print("Best model:\t", model)
    print("Desc. length:\t", model_len)
    plt.figure(figsize=(15, 5))
    plt.plot(desc_len)
    plt.xlabel("MCMC step", fontsize=14)
    plt.ylabel("Description length", fontsize=14)
    plt.title("MDL model: $%s$" % model.latex())
    plt.show()

run(pms, num_steps, thinning=100)

Parameters:

Name	Type	Description	Default
pms	Parallel	Parallel Machine Scientist (BMS is essentially a wrapper for pms)	required
num_steps	int	number of epochs / mcmc step & tree swap iterations	required
thinning	int	number of epochs between recording model loss to the trace	100

Returns:

Name	Type	Description
model	Tree	The equation which best describes the data
model_len	float	(defined as description length) loss function score
desc_len	List[float]	Record of loss function score over time

Source code in temp_dir/bms/src/autora/theorist/bms/utils.py

def run(
    pms: Parallel, num_steps: int, thinning: int = 100
) -> Tuple[Tree, float, List[float]]:
    """

    Args:
        pms: Parallel Machine Scientist (BMS is essentially a wrapper for pms)
        num_steps: number of epochs / mcmc step & tree swap iterations
        thinning: number of epochs between recording model loss to the trace

    Returns:
        model: The equation which best describes the data
        model_len: (defined as description length) loss function score
        desc_len: Record of loss function score over time

    """
    desc_len, model, model_len = [], pms.t1, np.inf
    for n in tqdm(range(num_steps)):
        pms.mcmc_step()
        pms.tree_swap()
        if num_steps % thinning == 0:  # sample less often if we thin more
            desc_len.append(pms.t1.E)  # Add the description length to the trace
        if pms.t1.E < model_len:  # Check if this is the MDL expression so far
            model, model_len = deepcopy(pms.t1), pms.t1.E
        _logger.debug("Finish iteration {}".format(n))
    return model, model_len, desc_len