"""Metrics for evaluation of the normalization."""
from __future__ import (nested_scopes, generators, division, absolute_import,
with_statement, print_function, unicode_literals)
from itertools import combinations
import numpy as np
from pandas import DataFrame, Series
from scipy.stats import ks_2samp
def check_vars(df, features):
"""Check if all variables in a list are present in a dataframe."""
features = Series(features)
is_feature_present = features.isin(df.columns)
missing_features_str = "Missing features: %s" % ', '.join(features[~is_feature_present])
assert is_feature_present.all(), ValueError(missing_features_str)
[docs]def ks_test_grid(df, features, sampling_variable='scanner'):
"""Calculate the Kolmogorov-Smirnov score for all pairs of scanners.
Parameters
----------
df: NDFrame of shape [n_subjects, n_features]
DataFrame with the subjects data.
features: list
List of the features to be considered on the Kolmogorov-Smirnov test.
sampling_variable: str, default='scanner'
Variable for which you want to group subjects.
Returns
-------
KS_by_variable: dict of NDFrames
Kolmogorov-Smirnov p-values to all pairs of instances in the sampling_variable column.
The keys in the dictionary are the variables in 'features'. The values of each entry are square NDFrames of
shape [n_vars, n_vars].
Raises
------
ValueError:
If the list of variables contain any variable that is not present in df.
Examples
--------
>>> ixi = DataSet('data/raw/IXI').data
>>> features = ['Left-Lateral-Ventricle', 'Left-Inf-Lat-Vent', ]
>>> KS = ks_test_grid(df, features, 'scanner')
>>> KS[features[0]]
+--------------------------+----------------------+------------------------+--------------------+
| | SCANNER01-SCANNER01 | SCANNER02-SCANNER01 | SCANNER03-SCANNER01|
+==========================++=====================+========================+====================+
|SCANNER01-SCANNER01 | NaN | NaN | NaN |
+--------------------------+----------------------+------------------------+--------------------+
|SCANNER02-SCANNER01 | 0.000759473 | NaN | NaN |
+--------------------------+----------------------+------------------------+--------------------+
|SCANNER03-SCANNER01 | 0.0539998 | 0.625887 | NaN |
+--------------------------+----------------------+------------------------+--------------------+
"""
check_vars(df, features)
groups = df.groupby(sampling_variable)
scanners_list = df[sampling_variable].unique()
scanners_list = scanners_list[np.argsort(df[sampling_variable].str.lower().unique())]
KS_by_variable = {}
for var in features:
KS = DataFrame([], index=scanners_list, columns=scanners_list)
for scanner_batch in np.array_split(np.array(list(combinations(scanners_list, 2))), 80):
for scanner_a, scanner_b in scanner_batch:
group_a = groups.get_group(scanner_a)[var]
group_b = groups.get_group(scanner_b)[var]
KS.loc[scanner_b][scanner_a] = ks_2samp(group_a, group_b).pvalue
KS_by_variable[var] = KS
return KS_by_variable