/LeNet5_CIFAR10

Master Thesis on Deep Reinforcement Learning

Primary LanguagePython

LeNet5_CIFAR10

Master Thesis project on Deep Reinforcement Learning.

Dependencies

  • tensorflow 2.3.1
  • tensorforce 0.6.2
  • numpy 1.19.4
  • matplotlib

How to use the agent

  • Step 1: Run tensorflow_train.py in order to train the agent.
  • Step 2: After training, go in tensorforce_train.py line 26 and change the weights file name with the one you got from Step 1.
  • Step 3: Check parameters in tensorforce_main.py (lines 17-23). Grid is set to be 2x2 by default.
  • Step 4: Run tensorforce_train.py. The agent should start to train indefinitely on a single random image from CIFAR10.

Acknowledgements

Thanks @SestoAle for providing the tracker_dense.py script.

Notes:

  • PyTorch net implementation is currently not used, but it's working.