#---------------------------------------------------------------------
# DeFCoM: A supervised learning genomic footprinter
# Copyright (C) 2016 Bryan Quach
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program. If not, see <http://www.gnu.org/licenses/>.
#----------------------------------------------------------------------
import uuid
import subprocess
from numpy import concatenate
from sklearn.metrics import roc_curve, auc
"""Methods for computing performance statistics."""
[docs]def pAUC(y_true, y_score, fpr_cutoff):
""" Calculate partial Area Under ROC (pAUC).
Computes a pAUC value given a specified false positive rate (FPR)
cutoff. It is important to note that the exact pAUC cannot be computed.
The accuracy of the calculation depends on the resolution of data
points produced by an intermediary ROC curve. The FPR data point
closest to and greater than the cutoff specified will be used for
interpolation to determine the pAUC at the specified FPR cutoff. For these
FPR values, the highest associated TPR values are used.
Args:
y_true: Array-like of true binary class labels in range {0, 1} or
{-1, 1} corresponding to y_score. The larger value represents
the positive class.
y_score: Array-like of target scores with higher scores indicating
more confidence in the positive class.
fpr_cutoff: A float specifying the FPR cutoff to use in computing
the pAUC. Must be in the interval (0,1).
Returns:
A float representing the pAUC value.
Raises:
AssertionError: The FPR cutoff is not in the interval (0,1)
"""
error_msg = "FPR cutoff must be in (0,1)"
assert fpr_cutoff > 0.0 and fpr_cutoff < 1.0
fpr, tpr, trash = roc_curve(y_true, y_score, drop_intermediate=False)
index_low = len([1 for i in fpr if i < fpr_cutoff])-1
index_high = index_low + 1
#Get interpolated TPR value from FPR cutoff
if index_low == -1: #No ROC data points lower than cutoff
x0 = fpr[0]
try:
x1 = min([xv for (c,xv) in zip([x>x0 for x in fpr],fpr) if c])
except ValueError:
x1 = x0
y0 = max([yv for (c,xv,yv) in zip([x==x0 for x in fpr],fpr,tpr) if c])
y1 = max([yv for (c,xv,yv) in zip([x==x1 for x in fpr],fpr,tpr) if c])
#Apply line derived from two closest points from FPR cutoff
tpr_cutoff = fpr_cutoff*((y1-y0)/(x1-x0)) + ((x1*y0-x0*y1)/(x1-x0))
#Segment full ROC to get partial ROC
fpr = [0.0] + [fpr_cutoff]
tpr = [0.0] + [tpr_cutoff]
elif index_high == len(fpr): #No ROC data points higher than cutoff
try:
x0 = max([xv for (c,xv) in zip([x<x1 for x in fpr],fpr) if c])
except ValueError:
x0 = x1
x1 = fpr[index_high-1]
y0 = max([yv for (c,xv,yv) in zip([x==x0 for x in fpr],fpr,tpr) if c])
y1 = max([yv for (c,xv,yv) in zip([x==x1 for x in fpr],fpr,tpr) if c])
#Apply line derived from two closest points from FPR cutoff
tpr_cutoff = fpr_cutoff*((y1-y0)/(x1-x0)) + ((x1*y0-x0*y1)/(x1-x0))
#Segment full ROC to get partial ROC
fpr = concatenate((fpr,[fpr_cutoff]), axis=0)
tpr = concatenate((tpr, [tpr_cutoff]), axis=0)
else:
x0 = fpr[index_low]
x1 = fpr[index_high]
y0 = max([yv for (c,xv,yv) in zip([x==x0 for x in fpr],fpr,tpr) if c])
y1 = max([yv for (c,xv,yv) in zip([x==x1 for x in fpr],fpr,tpr) if c])
#Apply line derived from two closest points from FPR cutoff
tpr_cutoff = fpr_cutoff*((y1-y0)/(x1-x0)) + ((x1*y0-x0*y1)/(x1-x0))
#Segment full ROC to get partial ROC
fpr = concatenate((fpr[:index_high], [fpr_cutoff]), axis=0)
tpr = concatenate((tpr[:index_high], [tpr_cutoff]), axis=0)
return auc(fpr,tpr)/fpr_cutoff