I noticed another bug in MultiCharts 12.0.16374.700 x64 (RC1), this time in Regular Optimization / Genetic Optimizer.
If I setup Genetic Optimizer to run in Basic mode, and set explicitly Population Size, Maximum Number of Populations settings, the optimizer seems to ignore these settings and always calculates some auto-generated number of iterations.
Here are the steps to repeat the issue:
0. I attached a test workspace to this post (see test.zip attachement). But it actually does not matter which signals / datafeeds to use. I've just throw out some random signals on the chart, and set optimization inputs to produce big number of combinations.
1. I do a Regular Optimization. Choose Genetic method, Basic Algorithm, and set Population Size: 500, Maximum Number of Populations: 50:
2. Run an optimization. Notice optimizer decides to calculate strategy 3252261 number of times, and optimization takes 17 hours with "Maximum Number of Populations: 50" setting.
3. Now, let's decrease Maximum Number of Populations to 10. I expect this to decrease calculation time approximately by a factor of 5, as now we have 50/10 == 5 times less populations to compute:
4. But the genetic optimizer still wants to calculate 3252261 iterations and Remaining Time is still expected to be 17-18 hours:
So GA optimizer ignores my settings? Please, investigate.
BUG: Genetic Optimizer ignores settings in Basic mode [SOLVED]
BUG: Genetic Optimizer ignores settings in Basic mode
- Attachments
-
- test.zip
- (16.44 KiB) Downloaded 156 times
- Svetlana MultiCharts
- Posts: 645
- Joined: 19 Oct 2017
- Has thanked: 3 times
- Been thanked: 163 times
Re: BUG: Genetic Optimizer ignores settings in Basic mode [SOLVED]
Hello, Sten,
Thanks for your feedback, it has been forwarded to the dev team and may be added to our to-do list for one of the future versions.
Thanks for your feedback, it has been forwarded to the dev team and may be added to our to-do list for one of the future versions.
Re: BUG: Genetic Optimizer ignores settings in Basic mode
Svetlana, thanks for forwarding my bug report to development team!
I want to make it clear what I describe above is a serious bug in Genetic Optimizer engine (or Optimization Wizard dialog). It should be investigated and fixed a soon as possible. This is not a feature request to be put on a to-do list for some future versions.
Please, keep me informed on development team response.
I want to make it clear what I describe above is a serious bug in Genetic Optimizer engine (or Optimization Wizard dialog). It should be investigated and fixed a soon as possible. This is not a feature request to be put on a to-do list for some future versions.
Please, keep me informed on development team response.
- Svetlana MultiCharts
- Posts: 645
- Joined: 19 Oct 2017
- Has thanked: 3 times
- Been thanked: 163 times
Re: BUG: Genetic Optimizer ignores settings in Basic mode
Sten,
I apologize if my reply was confusing. This issue is being analyzed by the dev team. I will keep you updated on the progress we have.
I apologize if my reply was confusing. This issue is being analyzed by the dev team. I will keep you updated on the progress we have.
Re: BUG: Genetic Optimizer ignores settings in Basic mode
I've also noticed an inconsistency in the incremental GAO:
On a 40 thread machine i setup a GAO that should be 4040 runs. But it runs as 4080: It appears to be because it is generating two parallel first generations.
So how does that work exactly?
From my understanding, gen1 should be one population of 40 backtests. gen2 should be 4 backtests of recombined children and so on. After gen1 it should only by 1000 rounds of 4 backtests (4x1000 +4 = 4040).
It looks like MC's process might be a little off here? It looks like another entire initial population of 40 is being created for some reason.
On a 40 thread machine i setup a GAO that should be 4040 runs. But it runs as 4080: It appears to be because it is generating two parallel first generations.
So how does that work exactly?
From my understanding, gen1 should be one population of 40 backtests. gen2 should be 4 backtests of recombined children and so on. After gen1 it should only by 1000 rounds of 4 backtests (4x1000 +4 = 4040).
It looks like MC's process might be a little off here? It looks like another entire initial population of 40 is being created for some reason.
- Svetlana MultiCharts
- Posts: 645
- Joined: 19 Oct 2017
- Has thanked: 3 times
- Been thanked: 163 times
Re: BUG: Genetic Optimizer ignores settings in Basic mode
Hello, wilkinsw,
This seems to be a different issue. We have forwarded it to our dev team. I will keep you updated as soon as any news is available.
This seems to be a different issue. We have forwarded it to our dev team. I will keep you updated as soon as any news is available.