Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

New optimization handles for outer scope users #201

Open
kasyanovse opened this issue Sep 21, 2023 · 0 comments
Open

New optimization handles for outer scope users #201

kasyanovse opened this issue Sep 21, 2023 · 0 comments
Labels
discuss Requires or would benefit from a discussion

Comments

@kasyanovse
Copy link
Collaborator

Problem description

FEDOT case.
Sometimes optimization cannot return appropriate results. For that case there are some parameters in FEDOT like num_of_generations, keep_n_best (it may be usefull for different structure of best graphs, but there is no way to check it), pop_size (it does not work as expected due to auto change of pop_size under GOLEM hood). As result only one with half parameters can adjust optimization process. It is very weak instruments for complex problem that cannot be solved by default GOLEM parameters.

More concise. I need to run some experiments:

  1. Optimization with high pop_size (1000-10000) and high amount of mutations (double, triple, etc). The only way for do it - clone GOLEM repo and hardcode what I need.
  2. Extend the initial population with the one mutation type and then run with all mutation types. Also clone repo and hardcode.
  3. Any another ideas.

Are there something that can help with it in current GOLEM release?

Problem definition

There is need for some simple instruments for tuning optimization.

Suggestion

The simplest and dumb idea is to create optimization pipelines.

  1. Pipeline for population optimization. It gets existing population and returns new population. Also there may be structure with any complexity.
  2. Pipeline for mutation. It gets the existing individual and return some new individuals.

Pros

  1. Optimization process may depend on task or on graph type, or any known parameter
  2. Optimization process may be changed by outer libraries like FEDOT for current domain
  3. Optimization may be optimized or created by metalearning algorithm (also by the outer libraries)
  4. Optimization may be very complicated and have a lot of conditional statements. Also different optimization approach may be combined (some parameters may be found by euristic algortihms as a substage of evolutional process - optimization will be faster)
  5. Optimization will become more clear for users and very handy for tuning

Cons

  1. It may be hard
  2. It may perturb the development plan of GOLEM
@kasyanovse kasyanovse added the discuss Requires or would benefit from a discussion label Sep 21, 2023
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
discuss Requires or would benefit from a discussion
Projects
None yet
Development

No branches or pull requests

1 participant