How to select a specific set of operators for the training about deep-symbolic-optimization HOT 4 CLOSED

Hi @sambatra,

You can explicitly set which operators you'd like to include in the "function_set" key of config.json under the "task" top-level key. For example:

"task": {
    "function_set" : ["add", "sub", "mul", "div"],
    ...
},
...

You can also include "const" as a token to if you'd like to include optimized floating-point constants, just keep in mind that this will slow the runtime down by a lot because it introduces an inner optimization loop.

I would like to avoid as much as possible as they make the results a bit harder to interpret.

As an alternative, with DSR you can add limits to the number of times a token can appear in a given expression. This allows you to still include trig/exp/log tokens, without letting them overwhelm the expression and become less interpretable. For example, you can limit each expression to having at most a total of 2 trig/exp/log terms by adjusting the config like this:

"task": {
    "function_set" : ["add", "sub", "mul", "div", "sin", "cos", "exp", "log"],
    ...
},
"prior" : {
    "repeat" : {"tokens" : ["exp", "log", "sin", "cos"], "max_" : 2},
    ...
    },
},
...

I would strongly recommend this alternative, because even just "max_" : 1 it will greatly increase the expressivity of your search space.

Let me know if this helps! Don't hesitate to reach out for anything else.

from deep-symbolic-optimization.

Hi @brendenpetersen

Many thanks for you help. I will experiment a bit.

from deep-symbolic-optimization.

Hi @brendenpetersen

It works pretty well, thanks!...I will try the alternative you suggested.

from deep-symbolic-optimization.

Wonderful! Stay tuned for an update to the code soon.

from deep-symbolic-optimization.