search

StochasticNumerics / mimclib

A software library for UQ methods
GNU General Public License v2.0
6 stars 8 forks source link

Very strange (large) time in column tT (tbl_lvls) #88

Closed litvinen closed 7 years ago

litvinen commented 8 years ago

Dear Abdul, In the column tbl_lvls.tT I see very large numbers, e.g., 111103 sec. and this is for levels 0|1 (iter id=639)

For this level I use relative coarse meshes with 1200 nodes. It should take max 10 minutes! It could be the case is Ml would be large. But Ml=7. For Ml=7 I expect to see maximum 1 hour (4K sec.), but not 100K.

Can you please check pictures below. See ow huge are tT!

screenshot from 2016-10-31 08 35 58 screenshot from 2016-10-31 08 36 08

haji-ali commented 8 years ago

The time tT is the totalTime (that is: time per sample times the number of samples)

litvinen commented 8 years ago

Yes, I understand. I just surprised that tT is so large, 91278 sec. for level 1.

haji-ali commented 8 years ago

The variance on level 0 is better than level 1. That's what you should be worried about. For tT, again, divide by number of samples to get a fair comparison between levels.

litvinen commented 7 years ago

@haji-ali >The variance on level 0 is better than level 1. That's what you should be worried about. Yes, I worry a lot about it. I observed many times that the mean El on level 1 is smaller as on level 2. I am almost sure that this is not a bug, but kind of aliasing (FFT) when I generate random perturbations on the sphere. Any ideas how to generate "better" input uncertainty ?

haji-ali commented 7 years ago

This really depends on the problem and I cannot possible give you advice without more understanding of the underlying model. In any case, this is not related to mimclib.