if e < 0.001:
break
return numpy.dot(UK, KI)
def mae(self, k):
snmf = numpy.array(self.matrix_factorization(k))
self.test_matrix = numpy.zeros(shape=(943, 1682))
result = 0
count = 0
for i in xrange(len(self.testdata_matrix)):
for j in xrange(len(self.testdata_matrix[i])):
if self.testdata_matrix[i][j] == 0:
continue
else:
self.test_matrix[i][j] = snmf[i][j]
count += 1
result += abs(self.test_matrix[i][j] -
self.testdata_matrix[i][j])
return (result / count)
if __name__ == '__main__':
nmf = NMF('/ml-100k/u.data')
for k in [10, 20, 30, 40, 50, 60, 70, 80]:
for m in [1, 2, 3, 4, 5, 6, 7, 8]:
for seed in [40, 60, 80, 100]:
nmf.shuffleData(m, seed)
nmf.matrix_factorization(k)
maerror += nmf.mae(k)
print 'The mean absolute error is:', float(maerror / (32 * 8))
