Discussion of article "Genetic Algorithms - It's Easy!" - page 10
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 on the upper loop it is assigned +1, so I subtract then -1 ...
That's where the dog is!!! And I think - how is it so - the article is already a year old - and everything works, nobody complains ... Thanks again for the article ... and for the explanations :)
@jooHello. Love your Zig-Zag, however it is slow on a large number of candles. Is it possible to get acceleration without loss of quality?
Sorry, but I am not the author of this indicator.
Please refer to the discussion page of this indicator.
I will not dispute the competence of the cited source, but I have to disagree.
The expediency of using GA in optimisation problems lies in reducing the number of FF runs required to determine the optimum compared to direct search.
If we follow the recommendation
then for a problem with 1000 arguments would require a population size of 11000 individuals! And that's 11000 FF runs on only 1 epoch! You can just as well use random gene generation and the result will not be much inferior in finding the optimum. The given source makes a "bet" that in a large population there will be enough genetic material to further develop the population towards improvement at each epoch. I am trying to achieve the same thing, but at the expense of playing probabilities with genetic operators without a total increase in FF runs.
Honestly, I couldn't imagine a problem with 1000 arguments. I use the algorithm to train a neural network, let's say five layers, although three is usually enough, so it turns out that the input is 14 variables for the five-layer, and 17 for the three-layer!!!! What can you put in there?!
Easy.
Here is the number of optimisable parameters for a 4-layer network with 2 hidden layers, for example for this one - 10-40-40-40-1:
10*40+40+40+40+40*40+40+40+40+40*1+1=2121 (weights of neurons and their shifts for 40+40+1=81 neurons).
As you can see, 2121 parameters need to be optimised for such a relatively small network.
Is it possible for you to post an example how you would use this library to optimize an Expert advisor like the way MT does?
It would be a great help for me
Thank you
Is it possible for you to post an example how you would use this library to optimize an Expert advisor like the way MT does?
It would be a great help for me
Thank you
I, unfortunately, there is no time for it, and the stimulus....
Maybe later.
Sorry, but I'm a bit dumb.
It turns out that in the algorithm with continuous genes considered here, such a parameter as a step is meaningless. We can only talk about accuracy if the result is known in advance and it is necessary to find arguments.
...
It turns out that in the algorithm with continuous genes considered here, such a parameter as step is meaningless. We can only talk about accuracy if the result is known in advance and it is necessary to find arguments.
Exactly right, if the distance between the genes is large enough then the step will be larger, as the genes get closer, the step decreases automatically, simply because the very concept of step is not used, but the accuracy with which you need to find a solution. The distance between genes is all the time divided by the random period, and it is constant 32768. The resulting number will be the conditional step.
If even the result is not known, you can assume the solution of what precision is required, it is easier than in unknown space to specify with what step you need to search for a solution.