Documentation scenario trial_memory_limit

[JosuaCarl]

question

Description

Set the limit for memory usage in a Scenario with trial_memory_limit = 8000

Steps/Code to Reproduce

scenario = Scenario(
configspace,
n_trials=1000,
min_budget=5,
max_budget=25,
n_workers=1,
output_directory=outdir,
walltime_limit=3600, # Max time in s (3600 = 1h)
cputime_limit=np.inf,
trial_memory_limit=8000,
)

smac = MultiFidelityFacade(
scenario,
train,
initial_design=initial_design,
intensifier=intensifier,
overwrite=True
)

incumbent = smac.optimize()

Expected Results

The Memory is limited to 8000 MB ~ 8 GB, as described in the docstring.

Actual Results

The memory seems to be limited to 8000 Bytes, the warning message
2024-03-20 11:17:27.320008: F tensorflow/tsl/platform/default/env.cc:74] Check failed: ret == 0 (11 vs. 0)Thread tf_numa_-1_Eigen creation via pthread_create() failed. Current memory usage in new process is 4571389952B but setting limit to 8000B. Likely to fail, try increasing the memory limit
and the resulting error pynisher.exceptions.PynisherException: Unknown reason for exitcode -6, no result or error recieved and killed process MLP.train(Configuration(values={...}), seed=..., budget=8.333333333333332) appears.

Versions

'2.0.2'

Proposed solution

• Change line 55 in smac/scenario.py from The maximum memory in MB that a trial is allowed to use. If not specified, to The maximum memory in Bytes (B) that a trial is allowed to use. If not specified,

[eddiebergman]

Just as an addition to this to the SMAC developers, the undeyling library allows for tuples of the form (8, "GB") or (8000, "MB") as well, which might be a bit more intuitive:
*https://github.com/automl/pynisher?tab=readme-ov-file#parameters