BOOTS-MBSAC

This is the implementation for the paper Bootstrapping the Expressivity with Model-based Planning.

Installment

# create a new environment if you want
# conda create -n <name> python=3.7
pip install -r requirements.txt
pip install -r lunzi/requirements.txt

To run a single experiment:

export PYTHONPATH=$PYTHONPATH:.
python examples/mb_sac.py --print_config --config configs/mbsac/walker2d.toml --log_dir /tmp/walker2d

You can find a TensorBoard event file at /tmp/walker2d and use the key episode/boots/reward/mean to generate your plot.

You can also use lunzi/run.py to run many experiments together:

./lunzi/run.py --n_jobs 10 --template 'python examples/mb_sac.py --print_config --log_dir /tmp/walker2d --config configs/mbsac/walker2d.toml'

MIT License.