How to communicate with engine easily outside stdin about minigo HOT 2 CLOSED

If you're trying to write a script, probably extending gtp.py is the place to do it.
serve.py is used for our front-end, minigui.

The gtp.py script is a straightforward way to load the model in the python engine.
If you need performance, you can build the gtp binary in cc/ and script inputs/outputs to that engine.

GTP is pretty strict about what it can output to stdout -- not much! Most of the useful output we put on stderr.

So you could:

1. write a script to run inferences with the model directly (lots of examples in oneoffs/)
2. write a script to interact with the engine via custom gtp command handlers (extending gtp.py)
3. write a script to interact with the engine via stdin and scrape stderr/stdout for the outputs

let me know if this helps!

from minigo.

Yes, it looks ok. I don't know why I wasn't able to understand it. I think the documentation could be better, there's a lot of scripts without a single word about usage. Thanks for pointing out oneoffs directory

from minigo.