Function main_plot_1D_attachment

hyperopt/plotting.py:176–257

Plots trail attachments, which are 1D-Data. A legend is only added if the number of plotted elements is < 10. :param trials: The trials object to gather the attachments from. :param attachment_name: Thename of the attachment to gather. :param do_show: If the plot should be sho

(
    trials,
    attachment_name,
    do_show=True,
    colorize_by_loss=True,
    max_darkness=0.5,
    num_trails=None,
    preprocessing_fn=lambda x: x,
    line_width=0.1,
)

def main_plot_1D_attachment(
    trials,
    attachment_name,
    do_show=True,
    colorize_by_loss=True,
    max_darkness=0.5,
    num_trails=None,
    preprocessing_fn=lambda x: x,
    line_width=0.1,
):
    """
    Plots trail attachments, which are 1D-Data.

    A legend is only added if the number of plotted elements is < 10.

    :param trials: The trials object to gather the attachments from.
    :param attachment_name: Thename of the attachment to gather.
    :param do_show: If the plot should be shown after creating it.
    :param colorize_by_loss: If the lines represening the trial data should be shaded by loss.
    :param max_darkness: The maximumg shading darkness (between 0 and 1). Implies colorize_by_loss=True
    :param num_trails: The number of trials to plot the attachment for. If none, all trials with a corresponding
    attachment are taken. If set to any integer value, the trials are sorted by loss and trials are selected in regular
    intervals for plotting. This ensures, that all possible outcomes are equally represented.
    :param preprocessing_fn: A preprocessing function to be appleid to the attachment before plotting.
    :param line_width: The width of the lines to be plotted.
    :return: None
    """
    # -- import here because file-level import is too early
    import matplotlib.pyplot as plt

    plt.title(attachment_name)

    lst = [l for l in trials.losses() if l is not None]
    min_loss = min(lst)
    max_loss = max(lst)

    if num_trails is None:
        plotted_trials = trials
    else:
        trials_by_loss = sorted(
            filter(lambda t: "loss" in t["result"], trials),
            key=lambda t: t["result"]["loss"],
        )
        plotted_trials = [
            trials_by_loss[i]
            for i in np.linspace(
                0, len(trials_by_loss), num_trails, endpoint=False, dtype=int
            )
        ]

    for trial in plotted_trials:
        t_attachments = trials.trial_attachments(trial)
        if attachment_name in t_attachments:
            attachment_data = np.squeeze(
                np.asanyarray(pickle.loads(t_attachments[attachment_name]))
            )
            if len(attachment_data.shape) == 1:
                attachment_data = preprocessing_fn(attachment_data)