Source code for datasets.brain.aggregate

import pandas as pd

from sacred import Experiment
from datasets.aggregate import meta_csv_aggregator_ingred
from utils import spans, hasflip, build_pairs, aggregate


brain_csv_aggregation = Experiment("brain_csv_aggregation",
    ingredients=[meta_csv_aggregator_ingred])


@brain_csv_aggregation.config
def config():
    meta_csv_file = "datasets/brain/test_data/meta_data_1200.csv"
    D = 3 # conversion window
    patient_id_key = "patient_id"
    builder = None

    protos = "HC_AD"

    pair_span = 2
    pair_slack = 0.25
    pair_low = pair_span - pair_slack
    pair_high = pair_span + pair_slack

    # Only t0 < t1 pairs
    if protos == "pairs":

        proto_csv_file = "datasets/brain/test_data/pair_protos.csv"

        def class_to_filter_fn(D):
            return {
                "HC": lambda x: x["HC"].all(),
                "AD": lambda x: x["AD"].all(),
                "MCI": lambda x: spans(x, "age", "at_most", D) and x["MCI"].all(),
            }

        class_to_filter = class_to_filter_fn(D)

        def builder_fn(pair_low, pair_high, patient_id_key):
            return lambda df: build_pairs(df,
                cond=lambda x,y: (
                    pair_low < y["age"].values - x["age"].values < pair_high
                ),
                no_duplicate=["class", patient_id_key])

        builder = builder_fn(pair_low, pair_high, patient_id_key)

    elif protos == "HC_AD": 

        proto_csv_file = "datasets/brain/test_data/HC_AD_protos.csv"

        class_to_filter = {
            "HC": lambda x: x["HC"].all(),
            "AD": lambda x: x["AD"].all()
        }

    elif protos == "spMCI":

        proto_csv_file = "datasets/brain/test_data/spMCI_protos.csv"

        def class_to_filter_fn(D):
            return {
                "sMCI": lambda x: spans(x, "age", "more_than", D) and x["MCI"].all(),
                "pMCI": lambda x: (
                            spans(x, "age", "more_than", D) and 
                            hasflip(x, sort_key="age", from_key="MCI", to_key="AD", within=D)
                        )
            }

        class_to_filter = class_to_filter_fn(D)

        builder = lambda df: df.ix[df["age"].idxmin()]

@brain_csv_aggregation.automain
def aggregate_protos(
    meta_csv_file,
    proto_csv_file,
    class_to_filter,
    builder,
    patient_id_key,
    _log):
    
    df = pd.read_csv(meta_csv_file)
                                 
    protos = aggregate(df, class_to_filter,
        group_key=patient_id_key,
        builder=builder)

    protos.to_csv(proto_csv_file, index=False)

    _log.info("n_samples = {}".format(protos.shape[0]))
    _log.info("n_unique_patients = {}".format(
        protos.groupby(patient_id_key).ngroups))