MDP-DynamicProg

Policy Iteration, Value Iteration and Prioritized Sweeping for simple grid world MDP control.

Grid World:

Rewards:

00  00 00 ... 00  00  00
00 -10 00 ... 00 -10  00
00  00 00 ... 00  00  00
          ...
00  00 00 ... 00  00  00
00  00 00 100 00  00  00
00  00 00 ... 00  00  00
          ...
00  00 00 ... 00  00  00
00 -10 00 ... 00 -10  00
00  00 00 ... 00  00  00

5 terminal states: 4 negative corners, 1 positive center

Actions:

Left(0) | Right(1) | Up(2) | Down(3)

Usage:

python main.py <algorithm> <OPTIONAL_FLAGS>

<OPTIONAL_FLAGS> are:

• --gamma (default 0.9, must be in (0.0, 1.0) )
• --width (default 5, must be odd integer in (5, 101) )

<algorithms> are:

• Policy Iteration algorithm: python main.py policy_iteration <OPTIONAL_FLAGS>
• Value Iteration algorithm: python main.py value_iteration <OPTIONAL_FLAGS>
• Prioritized Sweeping algorithm: python main.py prioritized_sweeping <OPTIONAL_FLAGS>