Skip to content

Latest commit

 

History

History
executable file
·
44 lines (32 loc) · 768 Bytes

README.md

File metadata and controls

executable file
·
44 lines (32 loc) · 768 Bytes

geometry_solver

geometry_solver is a deep reinforcement learning agent for geometry reasoning.

environment setup

First, you need to install python>=3.6. Then install dependancy using following comand:

pip install -r requirements.txt

Training

To train the DRL agnet, you can simply enter:

python rl_trainer.py

There are several options can be specified:

--algorithm: reinforcement algorithm.
--training_episode: Total training epoches.
--learning_rate
--gamma
--sample_num: number of samples taken by agent.
--test_num
--device: cpu or cuda
--show_process_bar
--log_interval
--epsilon
--target_replace_iter
--memory_capacity
--curriculum_learning: using curriculum learning

Evaluate

python run_test.py -- num 100