Action space

[aidan-curtis]

Task 1: Start taking action after initial ignition (random ignition/fixed ignition points)
Task 2: Can only take action before ignition (random ignition/fixed ignition points)
Task 3: Take actions before and after ignition ()

Actions:

• Change the type of vegetation (Certain brush size)
• Water (Certain brush size)

Option 1:
Network architecture: CNN --> (image -> discrete action space)
N teams - Start at fixed positions in a map
Choose to either to change vegetation/spray water/moveto

Option 2:
Network architecture: FCN (image-->image)
Action is an image that specifies fire and vegetation, but also

[williamshen-nz]

Will:

• C++ code doesn't terminate when weather data is finished (or randomize weather). Detect that simulation terminates
  • Sending done since it reaches max steps
• Randomizing ignition points, manipulate the CSV. Write it somewhere else.

Aidan

• Encode actions

Other Tasks:

• Tensorboard to visualize

Rewards:

• Fire
• Action costs
• No go actions
• Go into cities, pay infinite penalty

[williamshen-nz]

Additional Stuff from Slack:

• Get something to work for 20x20 orld
• Parallelize things so we can train
  • Write to separate CSVs so we can do that

[williamshen-nz]

Will

• Parallelized environments
• Fix the looping weather thing

Aidan

• Debug why RL isn't working. Maybe on 3x3 grid.

Next:

• Action space (x, y),
• Or output could be heatmap image coming out from FCNN
• More complex action space

[williamshen-nz]

Will's Tasks

• C++ code doesn't terminate when weather data is finished (or randomize weather). Detect that simulation terminates. Sending done since it reaches max steps
  • Added code in Cell2FireProcess to check whether line read from subprocess indicates end of simulation
• Randomizing ignition points, manipulate the CSV. Write it somewhere else.
  • Done in cell2fire/firehose/models.py
• Parallelized environments setup
  • We make a copy of the input directory that can be shared by all processes
  • Current logic in processes writes to different output directories on different process spawns
• Actually parallelize things with multiprocess, etc. To be done.

Should we loop on weather or just run fixed amount of steps up until weather files run out? Latter is easier and sufficient for now I think.

[williamshen-nz]

Will

• Debug the stupid issue
• Vectorize environment

Aidan

• Logging and monitoring
• Get the naive RL approach to work

[williamshen-nz]

Related to #5