python python-3.x conda biometrics python-bob

Python bob package API - how to format input data

I'm trying to use the bob package which is installed on my ubuntu using the Conda; however while using any of the measurement APIs like the bob.measure.eer_threshold the following error appears. I've prepared the data in 1D array but the error still exists. I've also tried to pass the pure 1D array to the function but it doesn't work. The error:

Traceback (most recent call last):   File "Test_bob.py", line 29, in <module>
threshold = bob.measure.eer_threshold(negatives, positives) ValueError: cannot convert `numpy.ndarray' which doesn't behave (memory contiguous, aligned, C-style, minimum 1 and up to 4 dimensions) into a `bob.blitz.array'

This is the code:

import bob
import bob.measure
import bob.blitz
import math
import numpy
from matplotlib import pyplot

fImpostor= open("Impostor.txt", "r")
fGenuine= open("Genuine.txt", "r")

positive_scores = []
negative_scores = []

for line in fImpostor:
    ImpScore = line.split()
    negative_scores.append(ImpScore[0])
fImpostor.close()

for line in fGenuine:
    GenScore = line.split()
    positive_scores.append(GenScore[0])
fGenuine.close()

positives = numpy.array(positive_scores)
negatives = numpy.array(negative_scores)

threshold = bob.measure.eer_threshold(negatives, positives)
FAR, FRR = bob.measure.eer_rocch(negatives, positives)

And this is the Genuine.txt file:

and the Impostor.txt:

please help me how to format and prepare data for such bob API methods.

Solution

That function expects a 1D array of floats, but you are actually reading your data in as strings. Additionally, instead of implementing the data-loading yourself, you could for instance use numpy.loadtxt to do this for you.