model failures
topepo opened this issue · 2 comments
I have had a few cases where it appears that the objective function is nan. This has occurred when too much regularization is used but, in the example below, there is no penalty applied. This uses RcppML 0.1.0.
library(RcppML)
library(modeldata)
library(Matrix)
data(biomass)
biomass <- biomass[, 3:7]
biomass <- as.matrix(biomass)
biomass <- t(biomass)
biomass <- Matrix(biomass, sparse = TRUE)
res <- nmf(biomass, k = 5, seed = 1)
#>
#> iter | tol
#> ---------------
#> 1 | nan
#> 2 | nan
#> 3 | nan
#> 4 | nan
#> 5 | nan
#> 6 | nan
#> 7 | nan
#> 8 | nan
#> 9 | nan
#> 10 | nan
#> 11 | nan
#> 12 | nan
#> 13 | nan
#> 14 | nan
#> 15 | nan
#> 16 | nan
#> 17 | nan
#> 18 | nan
#> 19 | nan
#> 20 | nan
#> 21 | nan
#> 22 | nan
#> 23 | nan
#> 24 | nan
#> 25 | nan
#> 26 | nan
#> 27 | nan
#> 28 | nan
#> 29 | nan
#> 30 | nan
#> 31 | nan
#> 32 | nan
#> 33 | nan
#> 34 | nan
#> 35 | nan
#> 36 | nan
#> 37 | nan
#> 38 | nan
#> 39 | nan
#> 40 | nan
#> 41 | nan
#> 42 | nan
#> 43 | nan
#> 44 | nan
#> 45 | nan
#> 46 | nan
#> 47 | nan
#> 48 | nan
#> 49 | nan
#> 50 | nan
#> 51 | nan
#> 52 | nan
#> 53 | nan
#> 54 | nan
#> 55 | nan
#> 56 | nan
#> 57 | nan
#> 58 | nan
#> 59 | nan
#> 60 | nan
#> 61 | nan
#> 62 | nan
#> 63 | nan
#> 64 | nan
#> 65 | nan
#> 66 | nan
#> 67 | nan
#> 68 | nan
#> 69 | nan
#> 70 | nan
#> 71 | nan
#> 72 | nan
#> 73 | nan
#> 74 | nan
#> 75 | nan
#> 76 | nan
#> 77 | nan
#> 78 | nan
#> 79 | nan
#> 80 | nan
#> 81 | nan
#> 82 | nan
#> 83 | nan
#> 84 | nan
#> 85 | nan
#> 86 | nan
#> 87 | nan
#> 88 | nan
#> 89 | nan
#> 90 | nan
#> 91 | nan
#> 92 | nan
#> 93 | nan
#> 94 | nan
#> 95 | nan
#> 96 | nan
#> 97 | nan
#> 98 | nan
#> 99 | nan
#> 100 | nan
all(is.na(res$w))
#> [1] TRUECreated on 2021-09-09 by the reprex package (v2.0.0)
@topepo Thanks for raising this issue, clearly it's causing you some trouble. Unfortunately, I could not reproduce your example using the dataset as supplied on two different machines (Windows and CentOS). I tried all seeds between 1 and 10000 for k = 5. In no case was all(is.na(res$w)) == TRUE.
Yes, too much regularization can drive a factor in w or h to complete sparsity, and thus numerical instability. That is the expectation.
Here's what my initial w matrix looks like:
> w_init <- nmf(biomass, k = 5, seed = 1, maxit = 0)$w
iter | tol
---------------
> w_init
[,1] [,2] [,3] [,4] [,5]
[1,] 0.2655087 0.89838968 0.2059746 0.4976992 0.9347052
[2,] 0.3721239 0.94467527 0.1765568 0.7176185 0.2121425
[3,] 0.5728534 0.66079779 0.6870228 0.9919061 0.6516738
[4,] 0.9082078 0.62911404 0.3841037 0.3800352 0.1255551
[5,] 0.2016819 0.06178627 0.7698414 0.7774452 0.2672207
> sessionInfo()
R version 4.0.3 (2020-10-10)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 10 x64 (build 19042)
I also tried on a different machine:
> sessionInfo()
R version 4.1.0 (2021-05-18)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: CentOS Linux 7 (Core)
> library(RcppML)
> library(modeldata)
> library(Matrix)
>
> data(biomass)
>
> biomass <- biomass[, 3:7]
> biomass <- as.matrix(biomass)
> biomass <- t(biomass)
> biomass <- Matrix(biomass, sparse = TRUE)
>
> res <- nmf(biomass, k = 5, seed = 1, maxit = 5)
iter | tol
---------------
1 | 4.58e-01
2 | 1.03e-01
3 | 5.50e-03
4 | 8.68e-04
5 | 6.20e-04
convergence not reached in 5 iterations
(actual tol = 6.20e-04, target tol = 1.00e-04)
> all(is.na(res$w))
[1] FALSE
I updated to the GH version and was able to reproduce what you have. Thanks!