Matrix contain inf or nan

[jastfkjg]

hello,

when I use mgpr to predict next observation, sometimes “matrix contains infs or nans" error occurs. I find it first appear in mgpr.py predict_given_factorizations, with code c = self.variance / tf.sqrt(tf.linalg.det(B)). It turns out that tf.linalg.det(B) is negative. What should I do about this ?

Thanks

[nrontsis]

Thanks for opening the issue. This shouldn't happen as B is by construction positive definite, thus its determinant must be positive. Perhaps the initial variance s that you are passing is not positive definite?

[jastfkjg]

Thanks. It's a prob in my code. I'll close this.