Open Bandd-k opened 8 years ago
If the MeasurementInterface implementation sets self.parallel_compile = True
, then the compile
method will be called multiple times in parallel based on the value of the parallelism flag, followed by run_precompiled
sequentially for each compiled version; run
is not called.
run_precompiled
is run sequentially under the assumption that running multiple trials in parallel will affect the fitness function, which is the case for OpenTuner's original application of program autotuning. If you're using OpenTuner for a general search problem where this doesn't hold, you can do the evaluation in compile
and just return the result from run_precompiled
(see how the Mario example does it). #24 is about implementing a MeasurementInterface subclass that does this automatically.
See also #69, which is also about documenting the parallelism flag.
Thank you!!
Hi! Could you explain how parallelism flag works please? I can't find differences between rosenbrock.py with parallelism and no. Thank you.