Optimisation and Out-of-Sample Testing. - page 11
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
But you still haven't answered the question of how big the sample should be :)
There is no definite answer, it depends on the specific TS. For me the size at which the results can be considered statistically significant is sufficient.
I've started to understand optimization and came across a lot of deadlocks(((
Objective: to write a program that would run sequentially all optimization options, and the results unloaded to Excel or DB.
To write a program that runs the tester from the command line is not a problem, but the results are all the same crap
1. it unloads into html without input parameters, the simplest way is deprived of the main link
2. impossible (or don't know) how to sort by number of operation, then it would be possible to recover input parameters by their order
3. how to copy results with input parameters manually to excel is known, but programmatically?
4. we could run the optimizer with a single combination of parameters, then unload it to html and then the parameters for this option will be known, then load the next combination, etc., but how much time would it take to call the tester? although it's a nice option, but very slow
5. I could also write a script that would download the results from the tester and run it programmatically at the end of testing, but I do not know if it is possible to run scripts in the tester and have access to test results
6. I could not find a decent solution in the forum, all some bits and boring attempts to circumvent restrictions, and by the way what are these restrictions associated with?? or lack of information how to circumvent them???
can someone smart tell me where to go? not for profit
If I do, I'll put it out there for public use.
I've started to understand optimization and came across a lot of deadlocks(((
Objective: to write a program that would run sequentially all optimization options, and the results unloaded to Excel or DB.
To write a program that runs the tester from the command line is not a problem, but the results are all the same crap
1. it unloads into html without input parameters, the simplest way is deprived of the main link
2. impossible (or don't know) how to sort by number of operation, then it would be possible to recover input parameters by their order
3. how to copy results with input parameters manually to excel is known, but programmatically?
4. we could run the optimizer with a single combination of parameters, then unload it to html and then the parameters for this option will be known, then load the next combination, etc., but how much time would it take to call the tester? although it's a nice option, but very slow
5. I could also write a script that would download the results from the tester and run it programmatically at the end of testing, but I do not know if it is possible to run scripts in the tester and have access to test results
6. I could not find a decent solution in the forum, all some bits and boring attempts to circumvent restrictions, and by the way what are these restrictions associated with?? or lack of information how to circumvent them???
can someone smart tell me where to go? not for profit
if i do, i'll post it for public use
'Automatic optimization of a trading robot in real trading'
'Testing and optimization management software'
Thank you
'Test & Optimisation Management Software'.
Working key price = 700p
'Automatic optimization of a trading robot in real trading'
I looked closely, it looks like there is some reading from html, I'll figure it out and check if it is good or not, but judging by the amount of text, the input parameters to the optimizer have to be solved through one place))
Thank you
'Test and Optimisation Management Software'.
Working key price = 700p
'Automatic optimization of a trading robot in real trading'
I looked closely, it looks like there is a reading from html, now I will figure it out and determine whether it is suitable or not, but judging by the amount of text to solve the problem of input parameters in the optimizer is through one place))
I do not know, I personally do not have enough skills and knowledge of programming to tie it all into a single bundle:
- optimization
- upload to a file
- get parameters from this file (not just one, but for each pass)
- write them into a file for downloading to the Expert Advisor for running on OOS..... and for several periods
- unloading again, so that primary parameters correspond to run results without shifts...........
- all this should work universally for any EA (with different quantity and different parameters) :)))))
The analysis of results is, probably, only by hand, as someone here said: "the most perfect neural network is the head".
In the meantime, I'm doing it manually (through "that place" called init), with some elements of automation. Last two weeks, I've been spending about 30 minutes preparing my computer/workstation (after turning it on) for all those operations. It's good to turn it off once a week, not more often.
But I think the results, both psychological and actual, are worth it.
1. in html unloaded without input parameters, the simplest path is deprived of the main link
They are there. When you hover over the run number, a hint with input parameters pops up. Since you are poking around in HTML, it shouldn't be too hard to get them out of there.
3. You know how to copy results with input parameters into excel manually, but programmatically?
Generate a CSV file with the required data and upload it to Excel programmatically.
6. I haven't found any decent solution in the forum, all these fragments and pathetic attempts to bypass the restrictions, and by the way, what are these restrictions related to?? or lack of information how to bypass them???
maybe someone clever will tell me where to go?
I haven't answered the rest of the questions already, as they are excluded by the correct answer to question 6. I think the correct answer was given earlier in this thread.
They are there. When you hover over the run number, a hint with the input parameters pops up
>> thank you. I'll get it out.
Optimization experts, please advise me, how to choose the best optimization horizon and history depth? I.e., for example, I need parameters obtained after the optimization on M1 to remain more or less stable for at least one week, and then I need to reoptimize them every week. For such an optimization horizon, what should be a sample of history for optimization - for a week, for a fortnight, for a month?
For the m1 and m5, I usually take a month to a month and a half story. This is usually enough for the "coming" week.
(I work with European indices - Dax, Futsi.)
For the n1 I take the history since last September. There is no sense to take earlier, as the pre-crisis market was "limp", and the tests of that time will not reflect the current realities.
I think it is reasonable to take 2-3 years on the n4 timeframe, or such a history, on which the Expert Advisor makes at least 250-300 deals.
Optimization experts, please advise me, how to choose the best optimization horizon and history depth? I.e., for example, I need parameters obtained after the optimization on M1 to remain more or less stable for at least one week, and then I need to reoptimize them every week. For such an optimization horizon, what should be the sample of history for optimization - for a week, for a fortnight, for a month?
The depth depends on the strategy used
--
how I optimize
getting for example
Some list, I choose parameters OPTIMIZABLE,
which is more common in profitable runs.
--
i.e. averaged