analysis_util.py

import pandas as pd

from generation_steps.metrics_analysis import correlations_for_group


def split_metrics_string(s):
    tokens = s.split("_")
    return tokens[1], tokens[3]


def get_ref_only_correlations_df(df):
    correlations_raw = correlations_for_group(df)

    idx = list(set("_".join(col.split("_")[:-1]) for col in correlations_raw.index))

    data = []
    for metrics in idx:
        data.append(
            {"metrics": metrics,
             "spearman": correlations_raw[f"{metrics}_spearman"],
             "pearson": correlations_raw[f"{metrics}_pearson"],
             }
        )

    result = pd.DataFrame.from_records(data=data, index="metrics").sort_index()
    result.index = pd.MultiIndex.from_tuples(result.index.map(split_metrics_string).tolist())
    result.index.set_names(["relative", "independent"], inplace=True)

    return result


def get_ref_only_correlations_for_groups(df):
    noref_correlations = {"all": get_ref_only_correlations_df(df)}

    for e2s in (False, True):
        for s2e in (False, True):
            suffix = ""
            if e2s:
                suffix += "+e2s"
            if s2e:
                suffix += "+s2e"
            if suffix == "":
                suffix = "golden"

            subdf = df[(df["end_to_start"] == e2s) & (df["start_to_end"] == s2e)]
            subdf_noref_corr = get_ref_only_correlations_df(subdf)
            noref_correlations[suffix] = subdf_noref_corr

    noref_correlations = pd.concat(noref_correlations, axis=1)
    return noref_correlations