play CartPole to 200 step with Q-learning using TensorFlow2.
this code is refined from Aurelien Geron's <Hands-on Machine Learning with Scikit-Learn, Keras, and TensorFlow>.
qlearning_cartpole_train.py:
train models ,write these models in "model-save-path" folder,default save a model every one step,so one can easier find a good model;
qlearning_cartpole_test.py:
test models ,read models from "model-save-path" folder ,test these model with several episodes ,write a result csv in "result-save-path"folder;
result_csv_and_good_models:
I run four times of the train and test code and record the result in csv.
several best model choosen from all result,all these model can play 200 steps;
curve_of_experiment_0.png:
max mean step curve of the experiment_0 (default parameters).One can plot a curve with the result csv of the policygradient_cartpole_test.py .
(1) make a folder to reserve the trained models # default : ./
(2) python3 qlearning_cartpole_train.py #default parameters is recommended
python3 qlearning_cartpole_test.py
(1) when testing , you can use "--render=True" to show the cartpole,but it will be slow.
(2) the model will saved every iteration(default),the example max-step curve shows below:
