ProBacktest extremely slow

Frankyboy · Sat, 09 Jan 2021 17:08:32 +0000

Dear all, I am experiencing extremely slowed ProBacktests lately (in both PRT11 and 10.3), especially for strategies that include MTF. Backtests (and I am not talking about WalkForward with multiple optimizations, just a simple regular backtest without any optimization, not even using tick-by-tick data), which have usually taken 10 seconds, now – for the past couple of days – take 5 Minutes to produce a result. Anybody experiencing the same situation? What is going on? Thanks in advance for advice.

This topic has 52 replies, 14 voices, and was last updated 3 years ago by robertogozzi.

Viewing 15 posts - 31 through 45 (of 53 total)

← 1 2 3 4 →

01/12/2021 at 2:25 PM #157554

Monochrome

Same problem as everyone here. Backtesting has become tediously slow. Never had this problem with 10.3.

Monochrome

Participant

Topics: 17

Replies: 116

Been thanked: 56 times

01/12/2021 at 4:29 PM #157575

Nicolas

PeterSt wrote:

I have one for you which takes 8 seconds in PRT-IB-V11. It takes 1:34 (one minute and a half) in PRT-IG-V11.

Interesting comment, thank you. The first thing that come to mind is the servers loads, since automatic trading is only possible with IG, most people are doing their backtests on IG accounts / servers. But, yes that’s almost 12 times longer, that’s why we need more examples and codes in order to precisely find where is the bottle neck and how we could make things better.

Nicolas

Keymaster

Topics: 81

Replies: 20863

Been thanked: 4773 times

01/13/2021 at 8:14 PM #157697

Frankyboy

The backtest speed seems to have improved by now. Attached is a trading system (the famous RSI2 by Larry Connors for the S&P – I programmed it for a 1h-chart, but it uses MTF to get the daily values for indicators – I had noticed that MTF strategies were slower than others). This strategy was extremely slow this past weekend (several minutes to load), still slow on Mon and Tues, but much faster today, almost as fast as it used to be in PRT10.3 times. I hope the speed issues got resolved and it stays like this also during weekends (as it used to be in PRT10.3 times).

Thanks again for your efforts, Nicolas!

1 user thanked author for this post.

Nicolas

Frankyboy

Participant

Topics: 1

Replies: 14

Been thanked: 5 times

01/14/2021 at 10:00 AM #157757

Nicolas

This is indeed a result of our investigations. Thanks also for your reports, we have found “small” things that can lead here and there to limit the backtests results to come faster or at least the same way that they did in previous version.

More improvements will come also after this first step, they need more kernel modifications of the backtest engine, but we should not wait too much I think.

2 users thanked author for this post.

Frankyboy,

GraHal

Nicolas

Keymaster

Topics: 81

Replies: 20863

Been thanked: 4773 times

01/17/2021 at 2:29 PM #158160

Monochrome

Backtesting taking over 10min…. Maybe it dosent do well with loop code?

Monochrome

Participant

Topics: 17

Replies: 116

Been thanked: 56 times

01/18/2021 at 11:18 AM #158255

Nicolas

@Monochrome

Please provide a sample code in order to understand why! 😉

Nicolas

Keymaster

Topics: 81

Replies: 20863

Been thanked: 4773 times

01/18/2021 at 2:38 PM #158288

Monochrome

I tested without this code, it loaded in seconds. With this code…minutes.

Dindex = ((exponentialaverage[10](close)-exponentialaverage[100](close))/exponentialaverage[100](close))*100

IF DINDEX CROSSES OVER 0 THEN
FLAG1 = BARINDEX
ENDIF
IF DINDEX CROSSES UNDER 0 THEN
FLAG1 = 0
ENDIF
IF DINDEX > 0 THEN
A1 = HIGHEST[max(1,BARINDEX-FLAG1)](DINDEX)
ELSE
A1 = 0
ENDIF

a20 = barindex-flag1
flag2 = 0
for a = 0 to a20-1
if dindex[a] = a1 then
flag2 = barindex[a]
break
endif
next

IF DINDEX > 0 THEN
A1A = LOWEST[max(1,BARINDEX-FLAG2)](DINDEX)
ELSE
A1A = 0
ENDIF

Dindex = ((exponentialaverage[10](close)-exponentialaverage[100](close))/exponentialaverage[100](close))*100

IF DINDEX CROSSES OVER 0 THEN

FLAG1 = BARINDEX

ENDIF

IF DINDEX CROSSES UNDER 0 THEN

FLAG1 = 0

ENDIF

IF DINDEX > 0 THEN

A1 = HIGHEST[max(1,BARINDEX-FLAG1)](DINDEX)

ELSE

A1 = 0

ENDIF

a20 = barindex-flag1

flag2 = 0

for a = 0 to a20-1

if dindex[a] = a1 then

flag2 = barindex[a]

break

endif

IF DINDEX > 0 THEN

A1A = LOWEST[max(1,BARINDEX-FLAG2)](DINDEX)

ELSE

A1A = 0

ENDIF

Hope its resolved soon.

Thank you

Monochrome

Participant

Topics: 17

Replies: 116

Been thanked: 56 times

02/02/2021 at 10:55 AM #160115

Nicolas

FYI, a new build has been pushed live yesterday for IG accounts, it should improve the backtests speed. I’d like to get your feels about it, could you please give here your feedbacks please? Thanks!

2 users thanked author for this post.

Midlanddave,

GraHal

Nicolas

Keymaster

Topics: 81

Replies: 20863

Been thanked: 4773 times

02/02/2021 at 1:56 PM #160142

Paul

Optimisation crashed…Started last night and was a heavy one but running strong. Come back the the computer 6 hours later and everything was gone. Lost my new created strategy.Made saves in between but still…not enough.

Paul

Participant

Topics: 40

Replies: 812

Been thanked: 660 times

02/02/2021 at 2:03 PM #160143

GraHal

Paul wrote:

Made saves in between but still…not enough.

Commiserations, but how did you (try) and save … so we all know how not to do it?

Backtest is at a standstill for me currently and I am only testing 40 combinations over 100k bars!

GraHal

Participant

Topics: 211

Replies: 8766

Been thanked: 2179 times

02/02/2021 at 2:15 PM #160147

nonetheless

I presume he just means Ctrl-S, which should save all your changes inc algos and optimizations … but of course it doesn’t. Export is the surest option, but that often causes crashes. Copying to Notepad has saved me a few times, but it’s a bore to have to do it.

Personally I find backtests to be running a bit faster today.

nonetheless

Participant

Topics: 83

Replies: 1237

Been thanked: 978 times

02/02/2021 at 2:16 PM #160148

GraHal

GraHal wrote:

Backtest is at a standstill for me currently

Mmm Quit PC No 1 and logged into PC No 2 and the 40 x 100k bars backtested ran in a few seconds (as I would expect).

I had completed about 10 or so backtests on PC No 1 and they were getting slower each new backtest I did.

My hypothesis is that processes are not being fully shut down / exited on the PRT server and so these hanging processes build up to some level which prevents further work!?

My workaround is going to be – File > Quit > Save and Exit – after every 3 backtest until the above hanging is no longer experienced.

GraHal

Participant

Topics: 211

Replies: 8766

Been thanked: 2179 times

02/02/2021 at 2:29 PM #160151

GraHal

nonetheless wrote:

Ctrl-S, which should save all your changes inc algos and optimizations … but of course it doesn’t

As a trial I entered fixed value = 34 into the Definition of Variables (DoV) window, closed the DoV window, closed the Code Editor, then did File > Quit > Exit without Saving.

I then logged back in and checked and the fixed value = 34 had been saved (even though I clicked Exit without Saving).

My hypothesis is that when the non-function limit (see my previous post) is reached then lots of PRT functions stop working correctly??

GraHal

Participant

Topics: 211

Replies: 8766

Been thanked: 2179 times

02/02/2021 at 2:43 PM #160155

GraHal

Ooops I omitted red text in above …

As a trial I entered fixed value = 34 into the Definition of Variables (DoV) window, closed the DoV window, closed the Code Editor, then did Ctrl+S, then did File > Quit > Exit without Saving.

GraHal

Participant

Topics: 211

Replies: 8766

Been thanked: 2179 times

02/02/2021 at 2:51 PM #160156

nonetheless

good to know ctrl-s works at least some of the time 😁

1 user thanked author for this post.

GraHal

nonetheless

Participant

Topics: 83

Replies: 1237

Been thanked: 978 times

Author

Posts

Viewing 15 posts - 31 through 45 (of 53 total)

← 1 2 3 4 →

Find exclusive trading pro-tools on