Let's try implement Reinforcement Learning algorithms with TensorFlow 2.0.
Find a way to install TensorFlow 2.0 Beta on your system.
Bellow is a simple method that works on any OS,
without messing up existing TensorFlow environment.
- Dependencies: Conda
- Using GPU:
$ conda create -n tf2 python cudatoolkit=10.0 cupti cudnn $ conda activate tf2 $ pip install -r requirements.txt $ python test/test_gpu.py
- Or not:
$ conda create -n tf2 python $ conda activate tf2 $ pip install -r requirements_nogpu.txt
- Simple Policy Gradient
$ python learnrl/simple_pg.py train \ --env_name CartPole-v0 \ --lr 0.025 \ --n_epochs 50 \ --batch_size 5000 \ --do_render 1