Closed seninp closed 5 years ago
I'm in 6% testing.
Can you try using garbage collector inside STOMP code?
I found a quick solution. I'm working on tweaking this for next release:
open stomp-par.R and change (line 116 approx):
per_work <- max(10, min(250, ceiling(num_queries / 100)))
with:
min_per_work <- 200
max_per_work <- 10000
plateaux_n_works <- 400
per_work <- max(min_per_work, min(max_per_work, ceiling(num_queries / plateaux_n_works)))
And
at line 144 approx, add next to:
pb$tick(per_work)
the function gc():
pb$tick(per_work)
gc()
Thank you for the quick response! I will try tomorrow -- once back to the office and will update. Thanks!
On Tue, Mar 5, 2019 at 2:40 PM Francisco Bischoff notifications@github.com wrote:
I found a quick solution. I'm working on tweaking this for next release:
open stomp-par.R and change (line 116 approx):
per_work <- max(10, min(250, ceiling(num_queries / 100)))
with:
min_per_work <- 200 max_per_work <- 10000 plateaux_n_works <- 400 per_work <- max(min_per_work, min(max_per_work, ceiling(num_queries / plateaux_n_works)))
And
at line 144 approx, add next to:
pb$tick(per_work)
the function gc():
pb$tick(per_work) gc()
[image: image] https://user-images.githubusercontent.com/984592/53842373-2e3ef400-3f97-11e9-8111-7d7e557fa49e.png
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/franzbischoff/tsmp/issues/38#issuecomment-469887656, or mute the thread https://github.com/notifications/unsubscribe-auth/ABplukuv7uBXquhiWokltCNMccR0WHGuks5vTvJbgaJpZM4bdiih .
-- Mahalo, Pavel.
The first trial (original code), after 11% it blew up my 32GB memory. After the changes, I'm at 53%, with the memory usage of 5-7GB, oscillating.
> res <- tsmp(CLEAN_House5$Appliance1[1:1000000], window_size = 1000, exclusion_zone = 1 / 4, verbose = 3, n_workers = 4)
Warming up parallel with 4 cores.
STOMP [================================] 100% at 22 it/s, elapsed: 13h, eta: 0s
Finished in 12.84 hours
The result, to save your time :-) data.zip
Describe the bug Following up the SCRIMP pub I'm trying to see structural features in the energy consumption series. Subsetting 8.xM points to 1M and running stomp with two threads ... task gets killed. What am I doing wrong?
To Reproduce Fresh Ubuntu 16.04 install, running the following code:
runs for few hours then gets killed by OS:
Environment