I wanted to test a strategy where I optimized my own ranking systems for buying and selling, and the only way I can achieve this here is as follows:
... use the main system only to sort the order of purchases (which stock is "best" if 20 pass your rules simultaneously).
Buy Rule: Rating("Kjops_System_AI") > 90
Sell Rule: Rating("Salgs_System_AI") < 80
Main Ranking: Set this to a neutral system, e.g., based only on the factor MktCap (RankType=Higher). This ensures that the simulation always buys the largest companies first among those approved by your unique buying system.
This is progressing very slowly. Is there anything that can be done to speed up such a process?
Could you walk through your workflow?
Are you trying to brute-force rule testing and develop the ranking system simultaneously?
P123 can correct me if I am wrong, but I believe the function Rating() runs much slower than the built in ranking, using rules Rank>< , because the built in ranker results are cached, and the Rating() results are not…
The idea is quite straightforward: I am trying to find a solution that allows me to optimize a ranking system for the buyer side and a separate one for the seller side, based on the assumption that entirely different stock criteria may be relevant for the buy than those relevant for the sell.
I normally use a genetic algorithm for the selection of criteria, along with a CMA-ES algorithm for weighting. In this case, I would run an optimization process on both of the two independent ranking systems simultaneously. However, I have a slight problem: the simulation itself becomes very slow when I use Rating("Kjops_System_AI") > 90.