Engines often lose on time

[thomasahle]

• Chess Tuning Tools version: 0.9.3
• Python version: 3.9
• Operating System: Linux

Description

I'm sorry this is not a very good issue report; but I've noticed that I get a lot of "Engine lost on time" warnings (5-10% of games) when using tune. This happens even though I use long time controls, like "60+1", and my engines never lose on time when I run them through cutechess-cli myself.

I'm wondering if communication with tune adds some extra move overhead somehow?
Is there something I can do to reduce this?
I could also add a "move overhead" parameter to my engine to let it know how much time it should expect to lose per move. But I can't as easily do that to the engines I'm testing against...

(I'm tuning sunfish, and I'm currently using pygone and sunfish as opponents.)

[ChrisWhittington]

Set timemargin=1000Sent from my iPhoneOn 6 Jan 2023, at 08:59, Thomas Dybdahl Ahle ***@***.***> wrote: Chess Tuning Tools version: 0.9.3 Python version: 3.9 Operating System: Linux Description I'm sorry this is not a very good issue report; but I've noticed that I get a lot of "Engine lost on time" warnings (5-10% of games) when using tune. This happens even though I use long time controls, like "60+1", and my engines never lose on time when I run them through cutechess-cli myself. I'm wondering if communication with tune adds some extra move overhead somehow? Is there something I can do to reduce this? I could also add a "move overhead" parameter to my engine to let it know how much time it should expect to lose per move. But I can't as easily do that to the engines I'm testing against... (I'm tuning sunfish, and I'm currently using pygone and sunfish as opponents.) —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[kiudee]

Thanks for reporting this. This is an interesting issue, since the code basically just opens a process to run cutechess-cli:

chess-tuning-tools/tune/local.py

Lines 910 to 917 in c1859b7

	with subprocess.Popen(
	string_array, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, text=True
	) as popen:
	if popen.stdout is not None:
	for line in iter(popen.stdout.readline, ""):
	yield line
	else:
	raise ValueError("No stdout found.")

I wonder whether there is some substantial overhead incurred due to parsing the log output here:

chess-tuning-tools/tune/cli.py

Lines 492 to 499 in 342e3d2

	for output_line in run_match(**match_settings):
	line = output_line.rstrip()
	is_debug = is_debug_log(line)
	if is_debug and verbose > 2:
	root_logger.debug(line)
	if not is_debug:
	out_exp.append(line)
	out_all.append(line)