After optimizing a system in approx 20 hours and it has gone 20% loading. I got this message “Warning: in order to clear temporary files, your connection will expire soon. You should save your modifications now and restart application. Attached a file.
So I guess this have been happening several time while night when im trying to optimize. Everytime i wake up in the morning the program is shut off. I lost several hours of work and I did manage to catch the message this time.
Even though i pressed exit later same message came with 0 minutes then it automatically closed. [MODERATION: DELETED BAD WORDS ; RESPECT THE POSTING RULES]
I was launching my prt using java mode through ig (as that is the best (fastest) way to optimize algos?) as you can choose > 8 GB ram.
Correct me if im wrong..
PRT is shut down every 18 hours (it was 12 up to a couple years ago) to clear some temp files.
You need to make your backtests shorter and I suggest that YOU shut it down before it does that itself.
I shut down, and immediately launch it again, twice a day, at around 09:00 and around 20:00, to make sure it’s within 18 hours.
Why would it be like that its 2020..
Well its hard when you have a long code and when 6000 combinations takes 1% per hour to load… So there is no other way around it?
So there is no other way around it?
Optimise with coarse / wide steps first then refine in to narrower steps after each optimiser run.
20 hours to optimise … I get impatient after a few minutes so I use the above method.
Or alternatively optimise 1 variable at a time or at most 2 at a time.
Use this box https://www.prorealtime.com/it/contatto to tell what you don’t like, what you like and what you suggest.
thanked this post
I use wide steps and trying to include about 6 variables.. It is a long code which takes time therefor it would be nice if the program didnt shut off every 18 hours atleast that you can have the option to not do it.
I think one reason it shuts down every 18 hours is to get people to think more carefully about how to optimize their codes without having to use 100,000 combinations. Most of the time, there are easier and faster ways. You will end up with a curve fit either way.
Normally, it is more important not to find “the best” of 6.000 combinations, but rather to see whether you get broad ranges of all variables where they work reasonably well. With only 100 out of 6000 combinations showing up in the end, you won’t see this.
Backtests are made server side, they can take longer if you are not alone in the queue. Try to reduce your combinations like it has been suggested.
I removed the bad words form your post, please be polite in your next posts.
How much are the backtest time depending on the computer performance or net speed capacity?
How much are the backtest time depending on the computer performance or net speed capacity?
Very little dependecy whatsoever
In my opinion … 4GB RAM and above helps prevent hanging / freezing.
PRT Servers do all the processing so user computers do not need to be latest models etc.
It dosent seem to process the calculation of indicators on the servers. Each indicator takes up a cpu core.
So if you use lots of indicators i would recommend lots of cores and ram.
Each indicator takes up a cpu core.
You talking about Indicators displayed on charts (not Indicators used in backtests)?
Yes indicators displayed on charts.
I have a slow cpu and on 200k it takes quite some time to load.
