Avoid Lapack routine dges in Mahalanobis distance

Asked Oct 08 '15 at 09:15

Active Oct 08 '15 at 09:29

Viewed 1,621 times

I get the following error while running the mahalanobis distance

Error in solve.default(cov, ...) : 
  Lapack routine dgesv: system is exactly singular: U[2,2] = 0

This is my code.

calibration <- 10
mean <- colMeans(x[1:calibration,])
sx <- cov(x[1:calibration,])
d <- mahalanobis(x[1:calibration,], mean, sx)

And this the data

1 0.7071068
0 0.0000000
2 1.4142136
2 1.4142136
0 0.0000000
0 0.0000000
2 1.4142136
1 0.7071068
1 0.7071068
0 0.0000000

How can I avoid this error?

edited Oct 08 '15 at 09:29

asked Oct 08 '15 at 09:15

1

What is `calibration`? – Oct 08 '15 at 09:16
It's value is equal 10 – Raúl Parada Medina Oct 08 '15 at 09:26
the problem is not the lapack routine, it is the covariance matrix of your data that is singular, which means the determinant is 0, hence the matrix is not invertible and you cannot calculate the Mahalanobis distance. – Mamoun Benghezal Oct 08 '15 at 09:28
Right, which condition can I insert to avoid this calculation? my script cannot go through this error – Raúl Parada Medina Oct 08 '15 at 09:30
1

See `try` function. `try(d <- mahalanobis(x[1:calibration,], mean, sx), silent = TRUE)`. – Oct 08 '15 at 09:32
2

Possible duplicate of [R solve:system is exactly singular](http://stackoverflow.com/questions/6572119/r-solvesystem-is-exactly-singular) – zx8754 Oct 08 '15 at 09:34
@Pascal it works. thanks – Raúl Parada Medina Oct 08 '15 at 09:59

0 Answers0