LITE Mode still looking for C library

[awil5504]

Hi,

I seem to be running into an issue in lite mode where despite having GPTUNE_LITE_MODE=1, and options['lite_mode'] = True GPTune looks for a compiled c library. I took a look at the lcm.py file and it looks for the lib_gptuneclcm library in the build directory (which I don't even have in this case since I've kept my lite and full installs separate). I've included the traceback for my error below.

I am running this on a slurm scheduler with the modules: module load SciPy-bundle/2021.10-foss-2021b module load netCDF-Fortran/4.5.3-gompi-2021b

where my objective fn in the tuner driver calls another driver that does a subprocess mpirun. I have tried running both gptune.SLA and gptune.MLA and get the same error. Happy to provide any more detail as needed

Andrew

File "/data/eart-tmm/sedm6691/GPTUNE_LITE/GPTune/examples/Sandbox/Bling_lite.py", line 186, in main() File "/data/eart-tmm/sedm6691/GPTUNE_LITE/GPTune/examples/Sandbox/Bling_lite.py", line 161, in main (data, modeler, stats) = gptune.SLA(NS=nrun, NS1=int(NS/2), Tgiven=giventask) File "/data/eart-tmm/sedm6691/GPTUNELITE/GPTune/GPTune/gptune.py", line 1024, in SLA (data, modeler, stats) = self.MLA(NS, NS1, 1, [Tgiven], T_sampleflag=[True], function_evaluations=None, s ource_function_evaluations=None, models_transfer=None, mfs=mfs) File "/data/eart-tmm/sedm6691/GPTUNELITE/GPTune/GPTune/gptune.py", line 932, in MLA modelers[o].train(data = tmpdata, kwargs) File "/data/eart-tmm/sedm6691/GPTUNE_LITE/GPTune/GPTune/model.py", line 430, in train return self.train_mpi(data, i_am_manager = True, restart_iters=list(range(kwargs['model_restarts'])), kw args) File "/data/eart-tmm/sedm6691/GPTUNE_LITE/GPTune/GPTune/model.py", line 435, in train_mpi from lcm import LCM File "/data/eart-tmm/sedm6691/GPTUNE_LITE/GPTune/GPTune/lcm.py", line 59, in raise Exception(f"Cannot find the lib_gptuneclcm library. Try to set env variable GPTUNE_INSTALL_PATH corr ectly.") Exception: Cannot find the lib_gptuneclcm library. Try to set env variable GPTUNE_INSTALL_PATH correctly.

[liuyangzhuan]

That's odd. Can you print out options before (data, modeler, stats) = gptune.SLA(NS=nrun, NS1=int(NS/2), Tgiven=giventask) and paste it here?

[awil5504]

Sure!

options = Options() options['lite_mode'] = True

options.validate(computer) # See GPTune docs p.90/116 on PDF, seem to always use in examples

options['verbose'] = False
options['distributed_memory_parallelism'] = False
options['shared_memory_parallelism'] = False 
options['objective_evaluation_parallelism'] = False

[awil5504]

Not sure what happened with formatting, but the options.validate(computer) is commented out in my code, everything else is active

[liuyangzhuan]

No, I'm asking you to do a print(options) before line (data, modeler, stats) = gptune.SLA(NS=nrun, NS1=int(NS/2), Tgiven=giventask) to see all the dictionary content of options.

But what you have just replied already revealed the problem. You need to call options.validate(computer=computer) before gt = GPTune(problem, computer=computer, data=data,... which will set options['model_class'] ='Model_GPy_LCM', in addition to a few other changes to make sure the c library is not used.

[hrluo]

Hi Andrew, Is your problem solved? Can you provide what Yang was suggesting?

[awil5504]

Hi! Apologies for the delay, this solved the problem!