Netlab Bug Tracker

This table should have the current status of known bugs. Please look check here first before contacting me about a new one.

1 Call to eigs in Netlab function eigdec uses incorrect syntax.  3.2  Fixed in 3.3
2 Call to sort in eigdec fails if complex eigenvalues are returned.  3.2  Fixed in 3.3
3 gsamp can occasionally give errors if eigenvalues of covariance matrix are not positive reals.  3.2  Fixed in 3.3
4 In the main help index there is a link to demkmean but the actual html file is demkmn1.
 3.2  Fixed in 3.3
5 In gmminit the replacement of small entries in the covariance matrix (line 73) uses wrong index.  3.2  Fixed in 3.3
6 In glmderiv gradient calculation does not work with a mask.  3.2  Fixed in 3.3
7 In mlpfwd output calculation with softmax does not prevent overflow as in glmfwd.  3.2  Fixed in 3.3
8 In gmm help text, some references are incorrect.
 3.2  Fixed in 3.3
9 In gmmpost treatment of rows in which all activations are zero is deficient. Fixed version warns and sets all posteriors to equal values.  3.2  Fixed in 3.3
10 In book canonical variates worked example, canvar did not normalise calculations of within and between class covariance.
 Book  Fixed 18/11/03 
11 In glmtrain Bayesian priors were not taken account of. Fixed version uses correct version of linear regression for scalar alpha and direct pseudo-inverse of Hessian for logistic and softmax outputs.
 3.2  Fixed in 3.3
12 In mlp functions from book direct connections worked example, actfn was used as a field instead of outfn. In addition, mlpbkp did not treat the case of multiple outputs correctly.
 Book  Fixed 05/02/04
13 In glminit test for correct network type was syntactically incorrect.
 3.2  Fixed in 3.3
14 gperr did not compute e2 correctly when the prior was not uniform for all parameters.
 3.2  Fixed in 3.3
15 gmmsamp did not sample correctly from a mixture of PPCA Gaussians.
 3.2  Fixed in 3.3
16 Optimisation and training algorithms give a warning message when they terminate because the maximum allowed number of iterations has been reached. Now the word "Warning" has been removed.
 3.2  Fixed in 3.3

Employable Graduates; Exploitable Research