Scratchpad
Recurring TODOs
Release
-
Run on benchmarks and check performance.
-
Check through code for accurate comments, stray debugs, formatting.
-
Beef up docstrings so that pydoc produces a useful description of the equations and logic.
-
Run pylint on all .py files.
-
Read through wiki.
-
Read through How It Works and update.
-
Update setup.py.
-
Create videos walking through code and operation.
-
Run benchmark one more time.
-
Commit.
-
Download and test.
-
Tag.
-
Release on pypi.
python setup.py sdist upload -r pypi
Advertise release:
- Google group
- Google+ groups
For next major release:
- Standalone video demo
- Doc for how to install and run
- Doc non-technical How it Works
- Doc technical How it Works
Long-term roadmap
Make captivating demos that get the attention of roboticists, data scientists, and non-technical audiences. Remove all friction possible to implement and run these.
For now, focus on incrementally evolving the catch world. Improve, complexify, and beautify.
Color palette R,G,B
These are all defined in core.tools.py
- copper shadow 25, 22, 20 (#191614)
- copper highlight 253, 249, 240 (#fdf9f0)
- light copper 242, 166, 108 (#f2a66c)
- copper 175,102,53 (#af6635)
- dark copper 132,73,36 (#844924)
- green 20, 120, 150 (#147896)
Unprioritized pool of TODOs
- Should actions be binary or real valued? How should output levels be calculated?
- Include more than one action output?
- Should actions have magnitude < 1? maybe not from deliberation, only from reaction (predictive action based on past patterns) Have automatic (non-deliberate) actions of any sort?
- Integrate with OpenAI Gym
