- These codes reproduces a project rejected by ICML using RL to learn a pivot in simplex method, it's a final project for the Algorithms in Large Data in PKU
The.npy files are the data saved by numpy using Python, they are the real data to test the model
The data file can be preprocessed by data.py and data2.py, according to the how the data is saved
SimplexFuntion.py provides all the functions needed to implement the simplex method
3 pivot rule: Dantzig, Steepest edge and the Bland's rules are available
main.py tests the methods on the real data
main2.py tests the methods on one special simulated case where Dantzig ourperforms Steepest edge
misc.py DeepSimDQN wrapsim are the files needed for training RL
misc.py provides all the necessary functions needed
DeepSimDQN.py provides the network parameter and class
wrapsim.py provides how the environment is organized in this case
to train the model using maindeep.py in cmd
running with --train 1 will train the model,
--weight can load existing model to train again
if no --train is provided, the file will test the existing model loaded
some other parameters included epsilon greedy, exploration times,... can also be revised and viewed