gemblerz/gomas

gomas's Issues

What to log...

We need to come up with some measurements we would like to gather from simulations. The measurements can be used / refined to get answers of the questions we have, and furthermore guide MAS toward better (more reasonable) behaviors.

SC2 protobuf API list

We need to know what APIs are available in SC2 for our simulation.

Plus, what are the properties in each message? We would need some examples on how to use them.

'Gather' action is continuous.

In SC2, probe can do their actions, but like 'Gather', some actions are continuous. It means that after one cycle to gather minerals, probe doesn't stop to act gathering.

• I think after doing the action, the agent should be idle state. Therefore, the gathering motion must be turned off.
• My thought is we can figure out its problem using queue_command. After every action, we can order the units to stop. Then, the agents can be idle state.

To figure out this problem, I think we need to modify Action, Agent, Core class. So, can anybody help me?

Traceback error in Windows

Traceback (most recent call last):
File "pingpong.py", line 59, in
observation.disable_fog = False
AttributeError: 'RequestObservation' object has no attribute 'disable_fog'

There is 'disable_fog' attribute in 'RequestObservation', but it cannot read the attribute and generates the error. The code works well if I just erase that line occuring the problem.

Simulation set-up description

Setting a simulation environment needs to be documented. That should include

• Version of any software tools
• Instruction of how to install the tools and configurations

It would be better if there is a single script that configures the environment.

Design Goal model

A goal should be described well enough so that agents can interpret its purpose, required actions, conditions, restrictions, and relationships with other goals (optional) in a format, which is agreed upon all the participants. A goal may have sub-goals indicating that sub-goals (or one of them) need to be attained in order to achieve the parent goal.

Logging strategy

In order to analyze events triggered in a simulation, a single, well-defined, logging mechanism is required. It not only gives an ability to see what is happening in the simulation, but also allow to track agent's behaviors upon events they perceived.

For now, logging module in python would serve well; just need a file that logs everything in time domain.

Also, we could use elastic search and kibana for data analysis. Refer to this

Design Agent model

A proactive agent seeking the most profitable action needs to be modeled. The model interacts with goal model as well as its internal sub models, e.g., knowledge model, communication model. To have heterogeneity in the group, individuals can have different knowledge, beliefs in their model.

zmq error: IPC doesn't work on Windows

Windows OS doesn't support the pipe named IPC, so If we want to support on Windows, We need to use other pipes such as TCP or make an exception for Windows.

Design goal model

The goal model needs to be designed in a way that the agent model can,

• understand how the goals are organized (e.g., goal 2-1 needs to be achieved in order to attain goal 2)
• know requirement of the goals (e.g., goal2-1 needs 100 unit amount of resources and one worker)
• format goals and send formatted goals to other agents (e.g., JSON would be a good start for this).

Please add more things for the goal model if you have.