Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Encountering NaN when fitting Vectors with CCIPCA #291

Open
hv10 opened this issue Sep 12, 2024 · 1 comment · May be fixed by #294
Open

Encountering NaN when fitting Vectors with CCIPCA #291

hv10 opened this issue Sep 12, 2024 · 1 comment · May be fixed by #294

Comments

@hv10
Copy link

hv10 commented Sep 12, 2024

Hi,

I am encountering the following issue when using CCIPCA:

When I call fit! repeatedly with the same input vector it can happen that the CCIPCA eigenvectors & -values become NaN.

Minimal Example:

a = [0.027667, 0.0428616, 0.57036, 0.382638, 68.4809, 24.4805, 230.786, 32.9694]
o = OnlineStats.CCIPCA(3, 8; l=3)
fit!(o, a)
display(o.U)

fit!(o,a)
display(o.U)

Leading to output:

8×3 Matrix{Float64}:
 0.000113292  0.0  0.0
 0.000175511  0.0  0.0
 0.00233553   0.0  0.0
 0.00156684   0.0  0.0
 0.280418     0.0  0.0
 0.100244     0.0  0.0
 0.945031     0.0  0.0
 0.135004     0.0  0.0
8×3 Matrix{Float64}:
 -0.000113292  NaN  0.0
 -0.000175511  NaN  0.0
 -0.00233553   NaN  0.0
 -0.00156684   NaN  0.0
 -0.280418     NaN  0.0
 -0.100244     NaN  0.0
 -0.945031     NaN  0.0
 -0.135004     NaN  0.0

From my limited experience I can see that the issue occurs when we divide by o.lambda within the fitting procedure.
Either at

o.U[:, i] = o.xi / o.lambda[i]

or
o.U[:, i] = o.v/o.lambda[i]

when the current eigenvalue is o.lambda[i]=0.0.

Is this something which is attributable to user error? If so how can I make sure to avoid it in the future?

@joshday
Copy link
Owner

joshday commented Sep 12, 2024

@robertfeldt

hv10 added a commit to hv10/OnlineStats.jl that referenced this issue Nov 22, 2024
@hv10 hv10 linked a pull request Nov 22, 2024 that will close this issue
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging a pull request may close this issue.

2 participants