Topics
Replies
andi21
17 May 2017, 19:45
RE:
andi21 said:
Please, Spotware - have you got any advice for this problem for me?
Is it possible that the work is correctly separated to be processed on each core (that works good), but the actual workload so the passes / bots which run in parallel in optimization is limited to 4 bots / workloads etc.?
Why do i have this idea?
Because on my other pc which has 4 cores it works over all cores and with 100%.
But on my new pc it works also over all 16 cores BUT only with 25% which would be 4 cores / bots / workloads.
Could that be the problem?
Has anyone else encountered this problem on a pc with 8 or 16 or more cores?
@andi21
andi21
16 May 2017, 17:20
Hi Spotware,
I have just tried one other SampleBot here from Bots-Section - it is the same problem (cpu consumption only at about 25%, but all cores are used correctly). So we can be sure that it does not depend on the code of my bot. It is something out of my scope.
Do you possibly have another idea?
@andi21
andi21
16 May 2017, 14:43
Hi Spotware,
i am refering to optimization (not backtesting - i know that in backtesting it only runs one bot instance and hence one core/main thread), so multiple bots run in parallel.
Like i said all cores are used, so that is fine. But overall it does not consume 100%, but only around 25%.
Hope you can help me with this.
Thanks.
@andi21
andi21
31 Mar 2017, 11:17
RE: RE:
trend_meanreversion said:
andi21 said:
Forgot something to ask:
do you have also an real account at icmarkets and if yes how long have been there with the real account and also are you satisfied with them?
Thanks in advance.
Been with them for more than a year now. No major issues but obviously they are like other brokers and will slow you down in terms of execution, login issues etc from time to time. But overall you to have choose the best among the beasts ! cTrader hasn't become that popular yet and most of the regulated brokers don't offer it and if they do , then terms aren't attractive ( spreads, commissions, products etc.. ). I have only 2 choices to be honest ICmarket or Pepperstone. Pepperstone charges much more in terms of commissions and IC is a little better in spreads + commissions.
Thanks for your answer. I asked, because i am currently choosing between FxPro and IC Markets (running a lot of optimization runs before demo forward testing).
I think i will give both a try running two demo accounts in parallel when i am ready to run forward testing and then i see which of them performs better. Of course, afterwards when i am going live it could be getting different results (more worse), but i hope it does not. Let's see and try :).
But overall you to have choose the best among the beasts
You are absolutely right - that is why i asked ;).
@andi21
andi21
23 Mar 2017, 17:32
I got new info to this topic:
now i also got results in optimization with 0 trades. And it is the same behaviour like i described:
use parameters from optimization with 0 trades in backtesting -> backtesting GOT trades and is running fine.
So strangely same parameters generates 0 trades in opmtization and 900 trades in backtesting (all on the same machine).
Has anybody an idea, perhaps it could be a bug?
@andi21
andi21
23 Mar 2017, 13:01
RE: RE:
trend_meanreversion said:
andi21 said:
Hello trend_meanreversion,
i am watching the results of your bot in myfxbook a time long.
The performance in my opinion is great!
I am working on a bot since several months (fulltime) incl. many many days of optimization ("sadly" i got many parameters).
I got quite good results in year 2016 (i optimized for that).
And yesterday i thought i am ready to go for forward testing and i runned 2014-2016. But strangely the year 2014 and 2015 performed not good (only even). So i thought i am getting crazy because the finish is getting further away...
I wanted to ask you if you can tell us what your bot is doing to get such great results (entry, sl, tp, exit etc.)? I mean it would be awesome if you could even share the sourcecode, but we cannot expect that because you did a lot of effort to get your bot like this.
Nevertheless no matter if you could provide some details to us or not (like i said i would understand that completely if not) good luck for your live going next month (risky martingale is running :) - but i did not see much risk in the results :) )!
Best Regards
Andi21
Hi Andi21,
First of all thanks for your words of appreciation ..it really helps and keeps me motivated !
My advice to you regarding martingale systems ( without revealing any secret sauce :P )
1) Put onus on entry levels..search for over-strech in market ( probably use multi time frames to confirm )
2) Grid spacing..try to be innovative here :) ..fixed , scaled version, ATR based etc etc..
3) Try different martingale position sizing ( there are hunderds.. linear, exponetial inreasing/decreasing, multiplier, doubling etc etc )
4) Try different TP schemes ( usually Martingale systems don't have SLs like mine and that's why i call it risky )..
I would suggest you to go through all of my posts/algo/comments which i have written in cTDN forums and i am sure you will be able to find something that will help you in some way or other.
I am already running live my Risky Martingale [ but usually quite conservative so lock in quite early ..like this week i locked in only 100$ even though risky martingale in demo is up 550$ :( ]
Most important thing for martingale systems is to understand associated risk which is 100% DD !! You can't avoid that , you can just delay it ! So make most of it till the time it hasn't seen that DD ,and that's where my stable version will help me. It has quite low probability of ruin ( but of-course can blow my account ). It has survived 3 years of tick data across multiple currency pairs which give me 'faith' in it .
Hope it helps and all the best !!
Thank you for the tips! I will look after it to try to improve my bot more and more.
Thanks and good luck!
Best Regards,
Andi21
@andi21
andi21
23 Mar 2017, 01:38
Hello trend_meanreversion,
i am watching the results of your bot in myfxbook a time long.
The performance in my opinion is great!
I am working on a bot since several months (fulltime) incl. many many days of optimization ("sadly" i got many parameters).
I got quite good results in year 2016 (i optimized for that).
And yesterday i thought i am ready to go for forward testing and i runned 2014-2016. But strangely the year 2014 and 2015 performed not good (only even). So i thought i am getting crazy because the finish is getting further away...
I wanted to ask you if you can tell us what your bot is doing to get such great results (entry, sl, tp, exit etc.)? I mean it would be awesome if you could even share the sourcecode, but we cannot expect that because you did a lot of effort to get your bot like this.
Nevertheless no matter if you could provide some details to us or not (like i said i would understand that completely if not) good luck for your live going next month (risky martingale is running :) - but i did not see much risk in the results :) )!
Best Regards
Andi21
@andi21
andi21
17 Jan 2017, 14:55
Hello f.mogyi,
i think your variable bAuth OR bWeekend is a nullable bool ( bool? ), but it does not have a value. So either you check that with bAuth.HasValue / bWeekend.HasValue before your access your variables or you change your logic in code (no nullables etc.).
Best Regards,
Andi21
@andi21
andi21
16 Jan 2017, 11:20
Thanks for your answer Team Spotware.
Meanwhile i have further investigated this problem and found out the source:
- the problem does only occur with timeframes, that opens on one day and closes on the other day, so for example hour, m15 etc. are fine, but the problem occurs with hour12, daily etc.
- the thing is the bars get cut into two pieces, so an example:
- Backtesting starts at 10.12.2016
- Hour12-Bar at last index from PAST data opens at 09.12.2016 22:00:00 and closes at the beginning of Backtesting start so at 09.12.2016 23:59:59, so the Hour12-Bar only contains data of these two hours
- On the first Tick on Backtesting start, so on 10.12.2016 00:00:00, the new data will NOT continue to fill the already started but not finished Hour12-bar from the PAST Data, it will falsely open a new Hour12-Bar which then only will contain the left 10 hours
So i think the previous bar from PAST data should be continued instead of that a new bar is opened.
@andi21
andi21
12 Jan 2017, 18:17
RE:
ColossusFX said:
Check the events tab, this is normally because a trade is still open
Thanks for your answer ColossusFx.
That is correct - Trade 7 is still open, but there is always one more Trade which is in this example the 8th one.
So for 7th Trade that is clear now, but why is always one more "ghost"-Trade visible in equity chart? Perhaps that only occurs if there is at least one trade still opened left.
That is not an important problem, but it is just curious.
Nevertheless thanks for your answer.
@andi21
andi21
17 May 2017, 22:54
RE:
kricka said:
I have also made a very simple quick & dirty (could be much better i know, but it is q&d) sample console app to prove the situation:
using System; using System.Threading.Tasks; namespace ConsoleApplication1 { class Program { static void Main(string[] args) { Console.WriteLine(Environment.ProcessorCount); if (true) // set to false for single core use { int zzz = Environment.ProcessorCount; zzz = 4; // Comment this out to use all processors Action[] z = new Action[zzz]; for (int xx = 0; xx < zzz; xx++) { z[xx] = Set; } Parallel.Invoke(z); } else { Set(); } Console.ReadKey(); } private static void Set() { long x = 5; for (long i = 0; i < 12147483647; i++) { if (x == 4) { throw new Exception(); } else { x++; } } } } }The program proved me (like i said in my first post, that it is neither a problem of my bot code, because i also tested another simple sample bot, nor it is a problem of my new pc, because bench of cpu-z also uses 100%), that 100% are possible without any problem:
-> if i use 1 core (if (false)) -> 7% are used (100 % / 16 cores = about 7%)
-> if i use processorcount (comment out the zzz = 4;) -> 100% are used
-> if i use FIX 4 cores (using zzz = 4): 25% are used ((100% / 16 cores) * 4 cores = about 25% -> like in calgo
so everything should be fine, but probably there is a problem in limiting the actual workload (see my post /forum/calgo-support/11535#7).
So all 16 cores are used, but the workload is only limited to 4 so the work of 4 is divided on 16 cores instead of a workload of 16 that is divided on 16 cores.
@andi21