Some pseudocode implementing a tedana decision tree node according to a proposed specification.

tedana_decision_tree_pseudocode.py

from textwrap import dedent
import logging

LGR = logging.getLogger(__name__)


def list_to_str(lst, joiner='and'):
    if len(lst) == 1:
        return lst[0]
    elif len(lst) == 2:
        return ' {} '.format(joiner).join(lst)
    else:
        str_ = ', '.join(lst[:-1]) + ', {} '.format(joiner) + lst[-1]
        return str_


def decision_tree_step(comptable, if_true='reject', if_false='no_change',
                       decide_comps=('unclassified', 'provisional_accept', 'provisional_reject'),
                       x_thresh=0.5, f_r2_clustermaps=None, return_metrics=False,
                       decision_tree_steps=None):
    metrics_used = ['metric']
    state_enum = 'decision_tree_step'
    config = {'decision_tree_step': {
        'metrics_used': metrics_used,
        'variables_used': {
            'if_true': if_true,
            'if_false': if_false,
            'decide_comps': decide_comps,
            'x_thresh': x_thresh,
            # don't want to store actual maps, but a bool describing whether
            # or not they were used at all could be good
            'f_r2_clustermaps': f_r2_clustermaps is not None,
            }
        }
    }
    # Actual decision step
    reduced_comptable = comptable.loc[comptable['classification'].isin(decide_comps)]
    classification_index = reduced_comptable['metric'] > x_thresh
    if if_true != 'no_change':
        reduced_comptable.loc[classification_index, 'classification'] = if_true
    if if_false != 'no_change':
        reduced_comptable.loc[~classification_index, 'classification'] = if_false

    comptable.loc[reduced_comptable.index, 'classification'] = reduced_comptable['classification']
    comptable.loc[reduced_comptable.index, 'state'] += (state_enum + ';')

    description = dedent"""\
    Components classified as {} were subselected for further classification
    using the "decision_tree_step" method, in which components with an x_thresh
    greater than {} are classified as "{}" and those with values below the
    threshold are classified as "{}".
    """.format(list_to_str(decide_comps, 'or'), x_thresh, if_true, if_false)
    LGR.info(description)

    decision_tree_steps.append(config)
    return comptable, metrics_used, description, decision_tree_steps


def metric_node(comptable, Br_R2_clmaps, F_R2_clmaps, metric_name='dice_FR2'):
    assert comptable.shape[0] == Br_R2_clmaps.shape[1] == F_R2_clmaps.shape[1]
    comptable[metric_name] = np.zeros(comptable.shape[0])
    for i_comp in comptable.index:
        comptable.loc[i_comp, 'dice_FR2'] = utils.dice(Br_R2_clmaps[:, i_comp],
                                                       F_R2_clmaps[:, i_comp])
    comptable.loc[np.isnan(comptable['dice_FR2']), 'dice_FR2'] = 0

    description = dedent"""\
    The "{}" metric was calculated as the Dice coefficient of masked,
    cluster-extent thresholded F-statistics and beta-value maps.
    """.format(metric_name)
    LGR.info(description)

    return comptable, description