#480 Update ISC test documentation with caveat about FPRs

Merged Samuel Nastase snastase
Showing 1 of 1 files from the diff.

@@ -668,10 +668,9 @@
Loading
668 668
    distribution of summary statistics. The p-value corresponds to either a
669 669
    'two-sided', 'left'-, or 'right'-sided (default) test, as specified by
670 670
    side. According to Chen et al., 2016, this is the preferred nonparametric
671 -
    approach for controlling false positive rates (FPR) for one-sample tests
672 -
    in the pairwise approach. The efficacy of this approach for controlling
673 -
    FPRs in the leave-one-out approach has not yet been systematically
674 -
    evaluated.
671 +
    approach for controlling false positive rates (FPRs) for one-sample tests
672 +
    in the pairwise approach. Note that the bootstrap hypothesis test may not
673 +
    strictly control FPRs in the leave-one-out approach.
675 674
676 675
    The implementation is based on the work in [Chen2016]_ and
677 676
    [HallWilson1991]_.
@@ -714,8 +713,8 @@
Loading
714 713
    p : float, p-value
715 714
        p-value based on bootstrap hypothesis test
716 715
717 -
    distribution : ndarray, bootstraps by voxels (optional)
718 -
        Bootstrap distribution if return_bootstrap=True
716 +
    distribution : ndarray, n_bootstraps by voxels
717 +
        Bootstrap distribution
719 718
720 719
    """
721 720
@@ -1088,8 +1087,8 @@
Loading
1088 1087
    The p-value corresponds to either a 'two-sided', 'left'-, or 'right'-sided
1089 1088
    (default) test, as specified by side. According to Chen et al., 2016,
1090 1089
    this is the preferred nonparametric approach for controlling false
1091 -
    positive rates (FPR) for two-sample tests. This approach may yield
1092 -
    inflated FPRs for one-sample tests.
1090 +
    positive rates (FPRs) for two-sample tests. Note that the permutation test
1091 +
    may not strictly control FPRs for one-sample tests.
1093 1092
1094 1093
    The implementation is based on the work in [Chen2016]_.
1095 1094
@@ -1124,8 +1123,8 @@
Loading
1124 1123
    p : float, p-value
1125 1124
        p-value based on permutation test
1126 1125
1127 -
    distribution : ndarray, permutations by voxels (optional)
1128 -
        Permutation distribution if return_bootstrap=True
1126 +
    distribution : ndarray, n_permutations by voxels
1127 +
        Permutation distribution
1129 1128
    """
1130 1129
1131 1130
    # Standardize structure of input data
@@ -1281,7 +1280,8 @@
Loading
1281 1280
    False. Returns the observed ISC and p-values, as well as the null
1282 1281
    distribution of ISCs computed on randomly time-shifted data. The p-value
1283 1282
    corresponds to either a 'two-sided', 'left'-, or 'right'-sided (default)
1284 -
    test, as specified by side.
1283 +
    test, as specified by side. Note that circular time-shift randomization
1284 +
    may not strictly control false positive rates (FPRs).
1285 1285
1286 1286
    The implementation is based on the work in [Kauppi2010]_ and
1287 1287
    [Kauppi2014]_.
@@ -1323,8 +1323,8 @@
Loading
1323 1323
    p : float, p-value
1324 1324
        p-value based on time-shifting randomization test
1325 1325
1326 -
    distribution : ndarray, time-shifts by voxels (optional)
1327 -
        Time-shifted null distribution if return_bootstrap=True
1326 +
    distribution : ndarray, n_shifts by voxels
1327 +
        Time-shifted null distribution
1328 1328
    """
1329 1329
1330 1330
    # Check response time series input format
@@ -1349,7 +1349,7 @@
Loading
1349 1349
        else:
1350 1350
            prng = np.random.RandomState(random_state)
1351 1351
1352 -
        # Get a random set of shifts based on number of TRs,
1352 +
        # Get a random set of shifts based on number of TRs
1353 1353
        shifts = prng.choice(np.arange(n_TRs), size=n_subjects,
1354 1354
                             replace=True)
1355 1355
@@ -1436,7 +1436,8 @@
Loading
1436 1436
    to False. Returns the observed ISC and p-values, as well as the null
1437 1437
    distribution of ISCs computed on phase-randomized data. The p-value
1438 1438
    corresponds to either a 'two-sided', 'left'-, or 'right'-sided (default)
1439 -
    test, as specified by side.
1439 +
    test, as specified by side. Note that phase randomization may not
1440 +
    strictly control false positive rates (FPRs).
1440 1441
1441 1442
    The implementation is based on the work in [Lerner2011]_ and
1442 1443
    [Simony2016]_.
@@ -1477,8 +1478,8 @@
Loading
1477 1478
    p : float, p-value
1478 1479
        p-value based on time-shifting randomization test
1479 1480
1480 -
    distribution : ndarray, time-shifts by voxels (optional)
1481 -
        Time-shifted null distribution if return_bootstrap=True
1481 +
    distribution : ndarray, n_shifts by voxels
1482 +
        Phase-shifted null distribution
1482 1483
    """
1483 1484
1484 1485
    # Check response time series input format

Everything is accounted for!

No changes detected that need to be reviewed.
What changes does Codecov check for?
Lines, not adjusted in diff, that have changed coverage data.
Files that introduced coverage data that had none before.
Files that have missing coverage data that once were tracked.
Files Coverage
brainiak 91.52%
Project Totals (31 files) 91.52%
Loading