Attempts at using RL in optimizations on Quantum Algorithms
Following is an example command, you can tune all the arguments to choose whatever benchmark, harware, search depth, etc. is needed. Use --train to train the model, don't if you want to just compile your circuits.
python -m qroute --dataset small --hardware qx20 --search 200 --train
The code in the qroute library looks as follows:
- algorithms - The combining methods, MCTS, Annealers, etc.
- models - The Neural networks that evaluate a state and the actions from that state, edit for new architectures.
- environment - Basic code managing the circuit and updates to the state, can be edited to add state representations.
- memory - Storing data for replay, linearly, or with Prioritized Experience Replay.
- visualizers - Checking if the result is correct, making videos and plots.
The rest of the package contains code for the paper, plotting stuff, illustrator files, etc.
Following is demonstrative video:
A presentation with details of the routing process are found here: https://docs.google.com/presentation/d/1Q-Y84ltoNbW15tKF_Nuh4MOcfkwyFIr6loAM7aGyJHk/edit?usp=sharing