def scorer_auc(y_true, y_pred, n_jobs=1):
from sklearn.preprocessing import LabelBinarizer
shape = y_pred.shape
if len(shape) == 1:
y_pred = y_pred[:, np.newaxis]
le = LabelBinarizer()
y_true = le.fit_transform(y_true)
assert_array_equal(np.unique(y_true), [0, 1])
_, _, auc = fast_mannwhitneyu(y_pred[np.where(y_true == 0)[0], ...],
y_pred[np.where(y_true == 1)[0], ...],
n_jobs=n_jobs)
return auc
def _default_analysis(X, y):
# from sklearn.metrics import roc_auc_score
from jr.stats import fast_mannwhitneyu
# Binary contrast
unique_y = np.unique(y)
# if two condition, can only return contrast
if len(y) == 2:
y = np.where(y == unique_y[0], 1, -1)
return np.mean(X * y[:, np.newaxis], axis=0)
elif len(unique_y) == 2:
# if two conditions but multiple trials, can return AUC
# auc = np.zeros_like(X[0])
_, _, auc = fast_mannwhitneyu(X[y == unique_y[0], ...],
X[y == unique_y[1], ...], n_jobs=1)
# for ii, x in enumerate(X.T):
# auc[ii] = roc_auc_score(y, np.copy(x))
return auc
# Linear regression:
elif len(unique_y) > 2:
return repeated_spearman(X, y)
else:
raise RuntimeError('Please specify a function for this kind of data')
def _default_analysis(X, y):
"""Aux. function to nested_analysis"""
# Binary contrast
unique_y = np.unique(y)
# if two condition, can only return contrast
if len(y) == 2:
y = np.where(y == unique_y[0], 1, -1)
# Tile Y to across X dimension without allocating memory
Y = tile_memory_free(y, X.shape[1:]) # FIXME tile memory free is now automatic in numpy # noqa
return np.mean(X * Y, axis=0)
elif len(unique_y) == 2:
# if two conditions but multiple trials, can return AUC
# auc = np.zeros_like(X[0])
_, _, auc = fast_mannwhitneyu(X[y == unique_y[0], ...],
X[y == unique_y[1], ...], n_jobs=1)
# for ii, x in enumerate(X.T):
# auc[ii] = roc_auc_score(y, np.copy(x))
return auc
# Linear regression:
elif len(unique_y) > 2:
return repeated_spearman(X, y)
else:
raise RuntimeError('Please specify a function for this kind of data')