[SHARE] GAB - Gekko Automated Backtests
#21
(03-26-2018, 07:36 AM)donkykong017 Wrote:  
2./3. yeah that seems to work. tried different strats and they all work. this definitely had something to do with the tulib talind indicators. but seems to be working great now.
4./6. yeah i got the sorting. i had some special cases where i tested other strats where i would like so take 2/3 values into account when sorting. but this should be solved when i can use more than 99 results. when i have 10000+ runs there should be lots more in consideration that could be very useful. instead with 99 there could be lots of results that never show e.g. when sorting with Strat profit there could be false positives (results with billions and trillions of profit) that i would like to avoid. a combination of Strat profit and sharpe could be a good idea.

5. ok got that. so is the default with backtesting the default from the UI (0.05)?

7. one thing that came to my mind with testing Strats that use Parameters in the 0.001 range: the average Strat parameters in the results view only shows whole numbers. more detailed views would be awesome there.

i will have to have a look at the new update though and will test there. 

thx for the update 
cheers 
DK

5. I do not know, i'm a version behind now so one task i got is to update Gekko and fix any errors that might occur. The last update took around 9 hours to complete, so i'm a little tired at the moment (worked on it from 18:00 > 04:00).

7. Yes, this is currently a problem. The average parameters for float values works better now in the last version but it isn't really 100% yet. The code is better in that it is more organized now so fixes should be a little more simple in the future.
  Reply
#22
(03-26-2018, 09:59 AM)tommiehansen Wrote: 5. I do not know, i'm a version behind now so one task i got is to update Gekko and fix any errors that might occur. The last update took around 9 hours to complete, so i'm a little tired at the moment (worked on it from 18:00 > 04:00).

7. Yes, this is currently a problem. The average parameters for float values works better now in the last version but it isn't really 100% yet. The code is better in that it is more organized now so fixes should be a little more simple in the future.

5. ok no stress man. take your time. and sleep  Big Grin

7. yeah great. it would be cool to have. but i have to get home today after work and check my backtests first anyways.

then i will have a look at the new update and work on from there with this version. will report back when something comes up.

thanks man
  Reply
#23
just to be shure. the new update loses all the fancy looks. its just plain text and pretty hard to use. i guess this is not intentional. 
also the select.php is gone so if it is intentional the readme is not correct anymore. you have to start with index.
  Reply
#24
Oh... Why don't you run your backtesting on your work computer?

This would be preferrable since your work computer probably is a desktop and could run 24/7 much, much... better then your laptop.
If you got a little $$$ laying around you could also buy a small box with some older desktop i5/i7 that still would be much faster then your laptop (due to the fact that it's desktop class...). You can basically find this on eBay or anywhere else for under $200.

To simply remote into it (and check stuff from home) use something simple such as TeamViewer that doesn't require you to become an expert on remote access security and stuff like that.

Running stuff 24/7 on a laptop isn't something i would recommend.
I use mine mainly for testing, not really for doing 2h+ runs...
  Reply
#25
(03-26-2018, 04:08 PM)donkykong017 Wrote: just to be shure. the new update loses all the fancy looks. its just plain text and pretty hard to use. i guess this is not intentional. 
also the select.php is gone so if it is intentional the readme is not correct anymore. you have to start with index.

1. No, try reloading it for real since it might have dropped links to etc (or your server might differ from mine thus it not finding stylesheets etc).
Edit: Come to think of it i hardcoded the basePath to /gab/ so if you got any other dir it'll break. This stuff happens when one codes at night, will fix @ future.

2. Yes, there is no 'select.php'; it has been renamed to index.php since 'select' will be the default start screen anyway.
  Reply
#26
In your screenshot with the insane results, the market also was performing really really well.
That is because you took a very long dataset to backtest on.

Do you do that for general robustness of the result?
My hunch is that this is a bad idea as the next time will have little to no resemblence to most of this time. Seeing these market increases is not realistic (otherwise we all should hodl).
  Reply
#27
(03-26-2018, 05:39 PM)simpsus Wrote: In your screenshot with the insane results, the market also was performing really really well.
That is because you took a very long dataset to backtest on.

Do you do that for general robustness of the result?
My hunch is that this is a bad idea as the next time will have little to no resemblence to most of this time. Seeing these market increases is not realistic (otherwise we all should hodl).

I fail to see how this is related to the tool. It shows result and cares less about backtesting theory, it's up to the user to interpret the results.
If you want to discuss backtesting theory and and hunches it may be a better idea to start a thread about that?

Just for you though -- here's a screen of something that is really over the top:
https://i.imgur.com/ErudFDX.png

Usually i don't share these sort of results though since it only leads to people believing all sorts of crazy stuff and because the simply fact that most doesn't seem to really grasp how to interpret results. That's the reason the main screenshot does not include such an insane result (but a lower 50k% gain). As i said -- it's up to the user to interpret the results. I only see it as 'these range of settings seem to stand a better change then these others'. How others interpret stuff or use/abuse the tool i leave up to them. Smile
  Reply
#28
I keep getting alot of runner.php curl errors when running on my RPI3. Threads are set to 1 and it seems that after each curl error (which does not stop the backtest in gekko ui) it starts the next backtest, leading to a lot of threads shooting the CPU to a steady 100% after some time.

Curl ERRORS are mostly status 0 or 500

Any idea how to prevent that? Is there some timeout? I've already increased php memory..

Thanks and BR,
Mathias
  Reply
#29
Well.. curl errors usually means it has timed out.
This is quite hard to debug unfortunately since you do not supply a way to recreate the issue?

Status 500 = internal server error which is quite a hard error.
Status 0 = Does not exist, this is an internal status meaning that cURL didn't return anything.

Timeout within PHP is set to 3600 meaning 60 minutes.
  Reply
#30
(03-26-2018, 09:24 PM)tommiehansen Wrote: Well.. curl errors usually means it has timed out.
This is quite hard to debug unfortunately since you do not supply a way to recreate the issue?

Status 500 = internal server error which is quite a hard error.
Status 0 = Does not exist, this is an internal status meaning that cURL didn't return anything.

Timeout within PHP is set to 3600 meaning 60 minutes.

Unless you have a RPI3 with Jessie and VNC access it would be hard to recreate Wink just hoped you have an idea. Not using local Firefox but the one on my remote PC reduced the errors, so it must have to do with performance, although it was not running above 60% CPU and 50% MEM when the errors occured.

I should avoid stressing my trading RPI anyway, so I will finally install gekko on my Windows 10 machine for backtesting purposes now.
  Reply


Forum Jump:


Users browsing this thread: