What's the maximum realistic number of permutations you can optimize in one shot using Exhaustive? I've found once I hit 2-3+ million it tends to freeze up towards the end of the run (after a day or so). 500,000 seems more manageable. Does that sound about right?
Could I run multiple optimization's of 500,000 if I run them using Non-Parallel Exhaustive? I'm just trying to handle this efficiently, as I have runs where I want to get through 10 million+ permutations. I've used PSO and the others, but upon backtesting them on strategies where I've already found my ideal permutation, they've always missed it. I tend to trust the brute force of Exhaustive optimization.
On similar note, what's the optimal computer setup for faster optimizations? I'm running an i9 10980HK + 32gb of Ram + SSD currently, but potentially want to upgrade. What would I look for in a newer setup?
Thank you!
Could I run multiple optimization's of 500,000 if I run them using Non-Parallel Exhaustive? I'm just trying to handle this efficiently, as I have runs where I want to get through 10 million+ permutations. I've used PSO and the others, but upon backtesting them on strategies where I've already found my ideal permutation, they've always missed it. I tend to trust the brute force of Exhaustive optimization.
On similar note, what's the optimal computer setup for faster optimizations? I'm running an i9 10980HK + 32gb of Ram + SSD currently, but potentially want to upgrade. What would I look for in a newer setup?
Thank you!
Rename
Take a look at the finantic.Optimizers extension. I bought it, and I regularly use the Particle Swarm and SMAC optimizers that the extension offers. Saves a ton of time.
https://www.wealth-lab.com/extension/detail/finantic.Optimizer
https://www.wealth-lab.com/extension/detail/finantic.Optimizer
Another optimization strategy is to use the Shrinking Window to find "the zone" to set up an Exhaustive optimization with fewer permutations. For example, if you knew the optimum for 2 parameters were around 20 and 50, you don't need to optimize from 10 to 100 for each one. You can go with something like 10 to 30 Step 2 for the first, and 30 to 70 Step 5 for the other.
It is not very useful to find the "optimum" on just one interval of backtest data. Always check with a second interval (and/or another portfolio)
You'll find out that you need a " stable region of parameter values" to get good results outside your Insample interval.
And the SMAC optimizer does not need more than 2000 iterations to find such a region.
No need for expensive hardware. Use clever software instead.
You'll find out that you need a " stable region of parameter values" to get good results outside your Insample interval.
And the SMAC optimizer does not need more than 2000 iterations to find such a region.
No need for expensive hardware. Use clever software instead.
Your Response
Post
Edit Post
Login is required