This is an easy TensorFlow implementation of Pensieve [1]. We trained Pensieve via PPO rather than A3C.
It's a stable version, which has already prepared the training set and the test set.
You can run the repo via typing:
python train.py
instead. Results will be evaluated on the test set (from HSDPA) every 300 epochs.
???
This model is the initial PPO model that is applied to the Pensieve problem.
The location of this model is src/rl_reference
and you can run the model in that folder as follows: python3 train.py
.
During the training process, we can leverage Tensorboard for monitoring current status.
python -m tensorboard.main --logdir [Results folder]
sat_data/
: contains the satellite tracesvideo_data/
: contains the sample videos' chunk size
- Dec. 28, 2022 (KJ Park): Summarized the codes and wrote the Readme.