Comments (6)
experimenting with always including best, issue still occurs. it appears
that gamma keeps increasing as no better solutions are found, and mutation
rate keeps increasing.
Now I think it will be helpful to run experiments without a dynamic mutation rate,
and without a dynamic gamma.
The population also averages their behaviors along one main cluster, it looks like.
See results in outs/ga-forage-daytime/gesmr-adapt-10kp-elite1
from evo.
this was apparently fine with gesmr-adapt-2kp. I don't remember what commit I ran it on, though.
from evo.
testing with no separate elite computation now
result: gamma collapse
from evo.
maybe need to do normal gaussian noise for updates? probably not
from evo.
x: half gamma on new best, commit d7e44ac on branch elite1
result: half gamma seems more stable, but new solutions are a negligible improvement on previous best. avg fitness plateaus at 1.5, mins at .5, maxes at 2.5 for most generations. This only uses 1m params
novelty isn't doing anything either, policies coverge on going to same corner after brief period of exploration
ideas:
more elites
more tests
larger grid + window sizes (increases model size too)
single agent
from evo.
todo: figure out how to launch visualization jobs of elites from julia
from evo.
Related Issues (20)
- perf: compression statistics
- todo: add env_pick and other_pick metric HOT 1
- bug: exchange still emerges somehow when playing against previous elite? HOT 1
- todo: confirm correctness of reversed-agent rollout
- idea: remove mutations as well as add
- todo: create metric to dictate strategy used for each agent
- todo: README
- todo: make eval!()
- todo: add massing integration test for 2 pops for a single gen
- todo: buffer output for trade env renderer HOT 1
- todo: change metric collector to collect individual rewards and update bc accordingly
- todo: make standalone program to generate image of trade env?
- todo: bc includes total fitness HOT 2
- bug: agents can place more than they pick
- bug: archive bc visualization can disappear in a later gen
- todo: rm pick/place counts on bc
- todo: make agents less adept at collecting a resource
- todo: reduce day length
- todo: profile update_pop! HOT 1
- todo: change create_next_pop back once tests are done
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from evo.