readme.md

shortest-paths-RL

Using reinforcement learning to find the shortest paths.

Requirements

• numpy
• networkx
• matplotlib
• imageio (optional, useful for generating the video/gif file to visualize)
• imageio-ffmpeg (optional, useful for generating the video/gif file to visualize)

To install them, try:

pip3 install numpy networkx matplotlib imageio imageio-ffmpeg 

Notes

• Please define the adjacent matrix for your problem. Note that the first row and the first column must correspond to the target state/node.
• You may need to modify the parameters of the reinforcement learning algorithms in order to solve your problem more effectively.

Examples

Here, we define the adjacent matrix as follows:

D = [[0, 4, 0, 0, 0, 0, 0, 8, 0],
     [4, 0, 8, 0, 0, 0, 0, 11, 0],
     [0, 8, 0, 7, 0, 4, 0, 0, 3],
     [0, 0, 7, 0, 9, 14, 0, 0, 0],
     [0, 0, 0, 9, 0, 10, 0, 0, 0],
     [0, 0, 4, 14, 10, 0, 3, 0, 0],
     [0, 0, 0, 0, 0, 3, 0, 3, 4],
     [8, 11, 0, 0, 0, 0, 3, 0, 5],
     [0, 0, 3, 0, 0, 0, 4, 5, 0]]

So the graph is as follows:

1. value iteration

To run value iteration algorithm, run:

python shortest_path.py -s vi

or

python shortest_path.py -s value_iteration 

The result is as follows:

2. policy iteration

To run policy iteration algorithm, run:

python shortest_path.py -s pi

or

python shortest_path.py -s policy_iteration 

The result is as follows:

3. Sarsa

The start node has been set to node 3 in the code. To run Sarsa algorithm, run:

python shortest_path.py -s sarsa

The result is as follows:

4. Sarsa(λ)

The start node has been set to node 3 in the code. To run Sarsa(λ) algorithm, run:

python shortest_path.py -s sarsa(lambda)

or

python shortest_path.py -s sarsa_lambda

The result is as follows:

5. q-learning

The start node has been set to node 3 in the code. To run q-learning algorithm, run:

python shortest_path.py -s q-learning

The result is as follows:

More details

More details can be seen in the code. You can also change the start node for Sarsa, Sarsa(λ) and q-learning algorithm.