def main():
Log.set_loglevel(logging.DEBUG)
modulename = "sample_ozone_posterior_average_slurm"
if not FileSystem.cmd_exists("sbatch"):
engine = SerialComputationEngine()
else:
johns_slurm_hack = "#SBATCH --partition=intel-ivy,wrkstn,compute"
johns_slurm_hack = "#SBATCH --partition=intel-ivy,compute"
folder = os.sep + os.sep.join(["nfs", "data3", "ucabhst", modulename])
batch_parameters = BatchClusterParameters(
foldername=folder,
max_walltime=24 * 60 * 60,
resubmit_on_timeout=False,
memory=3,
parameter_prefix=johns_slurm_hack)
engine = SlurmComputationEngine(batch_parameters,
check_interval=1,
do_clean_up=True)
prior = Gaussian(Sigma=eye(2) * 100)
num_estimates = 100
posterior = OzonePosteriorAverageEngine(computation_engine=engine,
num_estimates=num_estimates,
prior=prior)
posterior.logdet_method = "shogun_estimate"
proposal_cov = diag([4.000000000000000e-05, 1.072091680000000e+02])
mcmc_sampler = StandardMetropolis(posterior, scale=1.0, cov=proposal_cov)
start = asarray([-11.35, -13.1])
mcmc_params = MCMCParams(start=start, num_iterations=2000)
chain = MCMCChain(mcmc_sampler, mcmc_params)
chain.append_mcmc_output(StatisticsOutput(print_from=1, lag=1))
home = expanduser("~")
folder = os.sep.join([home, modulename])
store_chain_output = StoreChainOutput(folder)
chain.append_mcmc_output(store_chain_output)
loaded = store_chain_output.load_last_stored_chain()
if loaded is None:
logging.info("Running chain from scratch")
else:
logging.info("Running chain from iteration %d" % loaded.iteration)
chain = loaded
chain.run()
f = open(folder + os.sep + "final_chain", "w")
dump(chain, f)
f.close()
def test_shogun_on_serial_engine(self):
home = expanduser("~")
folder = os.sep.join([home, "unit_test_shogun_on_sge_dummy_result"])
try:
shutil.rmtree(folder)
except OSError:
pass
engine = SerialComputationEngine()
num_submissions = 3
sleep_times = randint(0, 3, num_submissions)
self.engine_tester(engine, sleep_times)
def main():
Log.set_loglevel(logging.DEBUG)
prior = Gaussian(Sigma=eye(2) * 100)
num_estimates = 2
home = expanduser("~")
folder = os.sep.join([home, "sample_ozone_posterior_rr_sge"])
computation_engine = SerialComputationEngine()
rr_instance = RussianRoulette(1e-3, block_size=10)
posterior = OzonePosteriorRREngine(rr_instance=rr_instance,
computation_engine=computation_engine,
num_estimates=num_estimates,
prior=prior)
posterior.logdet_method = "shogun_estimate"
proposal_cov = diag([4.000000000000000e-05, 1.072091680000000e+02])
mcmc_sampler = StandardMetropolis(posterior, scale=1.0, cov=proposal_cov)
start = asarray([-11.35, -13.1])
mcmc_params = MCMCParams(start=start, num_iterations=200)
chain = MCMCChain(mcmc_sampler, mcmc_params)
# chain.append_mcmc_output(PlottingOutput(None, plot_from=1, lag=1))
chain.append_mcmc_output(StatisticsOutput(print_from=1, lag=1))
store_chain_output = StoreChainOutput(folder, lag=50)
chain.append_mcmc_output(store_chain_output)
loaded = store_chain_output.load_last_stored_chain()
if loaded is None:
logging.info("Running chain from scratch")
else:
logging.info("Running chain from iteration %d" % loaded.iteration)
chain = loaded
chain.run()
f = open(folder + os.sep + "final_chain", "w")
dump(chain, f)
f.close()
return job
if __name__ == "__main__":
logger.setLevel(10)
num_repetitions = 10
# plain MCMC parameters, plan is to use every 200th sample
thin_step = 1
num_iterations = 5200
num_warmup = 200
compute_local = False
if not FileSystem.cmd_exists("sbatch") or compute_local:
engine = SerialComputationEngine()
else:
johns_slurm_hack = "#SBATCH --partition=intel-ivy,wrkstn,compute"
folder = os.sep + os.sep.join(["nfs", "data3", "ucabhst", modulename])
batch_parameters = BatchClusterParameters(
foldername=folder,
resubmit_on_timeout=False,
parameter_prefix=johns_slurm_hack)
engine = SlurmComputationEngine(batch_parameters,
check_interval=1,
do_clean_up=True)
engine.max_jobs_in_queue = 1000
engine.store_fire_and_forget = True
aggs = []
No aggregators are stored and results can be picked up from disc when ready.
This script also illustrates a typical use case in scientific computing:
Run the same function with different parameters a certain number of times.
Make sure to read the minimal example first.
"""
Log.set_loglevel(10)
# filename of the result database
home = expanduser("~")
foldername = os.path.join(home, "test")
db_fname = os.path.join(foldername, "test.txt")
batch_parameters = BatchClusterParameters(foldername=foldername)
engine = SerialComputationEngine()
# engine = SlurmComputationEngine(batch_parameters)
# here are some example parameters for jobs
# we here create all combinations and then shuffle them
# this randomizes the runs over the parameter space
params_x = np.linspace(-3, 3, num=25)
params_y = np.linspace(-2, 2, num=12)
all_parameters = itertools.product(params_x, params_y)
all_parameters = list(all_parameters)
shuffle(all_parameters)
print "Number of parameter combinations:", len(all_parameters)
for params in all_parameters[:len(all_parameters) / 300]:
x = params[0]
y = params[1]
No aggregators are stored and results can be picked up from disc when ready.
This script also illustrates a typical use case in scientific computing:
Run the same function with different parameters a certain number of times.
Make sure to read the minimal example first.
"""
Log.set_loglevel(10)
# filename of the result database
home = expanduser("~")
foldername = os.path.join(home, "test")
db_fname = os.path.join(foldername, "test.txt")
batch_parameters = BatchClusterParameters(foldername=foldername)
engine = SerialComputationEngine()
# engine = SlurmComputationEngine(batch_parameters)
# here are some example parameters for jobs
# we here create all combinations and then shuffle them
# this randomizes the runs over the parameter space
params_x = np.linspace(-3, 3, num=25)
params_y = np.linspace(-2, 2, num=12)
all_parameters = itertools.product(params_x, params_y)
all_parameters = list(all_parameters)
shuffle(all_parameters)
print "Number of parameter combinations:", len(all_parameters)
for params in all_parameters[:len(all_parameters) / 300]:
x = params[0]
y = params[1]
def test_serial_engine(self):
num_submissions = 3
sleep_times = randint(0, 3, num_submissions)
self.engine_helper(SerialComputationEngine(), sleep_times)
this script, we can collect results from the cluster and potentially
submit more jobs.
"""
Log.set_loglevel(10)
# oflder for all job files
home = expanduser("~")
foldername = os.sep.join([home, "minimal_example"])
# parameters for the cluster (folder, name, etcI
batch_parameters = BatchClusterParameters(foldername=foldername)
# engine is the objects that jobs are submitted to
# there are implementations for different batch cluster systems
# the serial one runs everything locally
engine = SerialComputationEngine()
# engine = SGEComputationEngine(batch_parameters)
# engine = SlurmComputationEngine(batch_parameters)
# On submission, the engine returns aggregators that can be
# used to retreive results after potentially doing postprocessing
returned_aggregators = []
for i in range(3):
job = MyJob(ScalarResultAggregator())
agg = engine.submit_job(job)
returned_aggregators.append(agg)
# This call blocks until all jobs are finished (magic happens here)
logger.info("Waiting for all jobs to be completed.")
engine.wait_for_all()
num_warmup = 500
thin_step = 1
num_iterations = 2000 + num_warmup
num_iterations = 100
num_warmup = 0
# hmc parameters
num_steps_min = 10
num_steps_max = 100
sigma_p = 1.
momentum_seed = np.random.randint(time.time())
compute_local = False
if not FileSystem.cmd_exists("sbatch") or compute_local:
engine = SerialComputationEngine()
else:
johns_slurm_hack = "#SBATCH --partition=intel-ivy,wrkstn,compute"
folder = os.sep + os.sep.join(["nfs", "data3", "ucabhst", modulename])
batch_parameters = BatchClusterParameters(
foldername=folder,
resubmit_on_timeout=False,
parameter_prefix=johns_slurm_hack)
engine = SlurmComputationEngine(batch_parameters,
check_interval=1,
do_clean_up=True)
engine.max_jobs_in_queue = 1000
engine.store_fire_and_forget = True
aggs_hmc_kmc = {}
def run_problem(prob_label):
"""Run the experiment"""
# /////// submit jobs //////////
# create folder name string
#result_folder = glo.result_folder()
from kcgof.config import get_default_config
config = get_default_config()
tmp_dir = config['ex_scratch_path']
foldername = os.path.join(tmp_dir, 'kcgof_slurm', 'e%d' % ex)
logger.info("Setting engine folder to %s" % foldername)
# create parameter instance that is needed for any batch computation engine
logger.info("Creating batch parameter instance")
batch_parameters = BatchClusterParameters(foldername=foldername,
job_name_base="e%d_" % ex,
parameter_prefix="")
use_cluster = glo._get_key_from_default_config('ex_use_slurm_cluster')
if use_cluster:
# use a Slurm cluster
partitions = config['ex_slurm_partitions']
if partitions is None:
engine = SlurmComputationEngine(batch_parameters)
else:
engine = SlurmComputationEngine(batch_parameters,
partition=partitions)
else:
# Use the following line if Slurm queue is not used.
engine = SerialComputationEngine()
n_methods = len(method_funcs)
# problem setting
ns, p, rx, cs = get_ns_model_source(prob_label)
# repetitions x len(ns) x #methods
aggregators = np.empty((reps, len(ns), n_methods), dtype=object)
for r in range(reps):
for ni, n in enumerate(ns):
for mi, f in enumerate(method_funcs):
# name used to save the result
func_name = f.__name__
fname = '%s-%s-n%d_r%d_a%.3f.p' \
%(prob_label, func_name, n, r, alpha,)
if not is_rerun and glo.ex_file_exists(ex, prob_label, fname):
logger.info('%s exists. Load and return.' % fname)
job_result = glo.ex_load_result(ex, prob_label, fname)
sra = SingleResultAggregator()
sra.submit_result(SingleResult(job_result))
aggregators[r, ni, mi] = sra
else:
# result not exists or rerun
job = Ex1Job(SingleResultAggregator(), prob_label, r, f, n)
agg = engine.submit_job(job)
aggregators[r, ni, mi] = agg
# let the engine finish its business
logger.info("Wait for all call in engine")
engine.wait_for_all()
# ////// collect the results ///////////
logger.info("Collecting results")
job_results = np.empty((reps, len(ns), n_methods), dtype=object)
for r in range(reps):
for ni, n in enumerate(ns):
for mi, f in enumerate(method_funcs):
logger.info("Collecting result (%s, r=%d, n=%d)" %
(f.__name__, r, n))
# let the aggregator finalize things
aggregators[r, ni, mi].finalize()
# aggregators[i].get_final_result() returns a SingleResult instance,
# which we need to extract the actual result
job_result = aggregators[r, ni, mi].get_final_result().result
job_results[r, ni, mi] = job_result
#func_names = [f.__name__ for f in method_funcs]
#func2labels = exglobal.get_func2label_map()
#method_labels = [func2labels[f] for f in func_names if f in func2labels]
# save results
results = {
'job_results': job_results,
# 'p': p,
# 'cond_source': cs,
'alpha': alpha,
'repeats': reps,
'ns': ns,
'method_funcs': method_funcs,
'prob_label': prob_label,
}
# class name
fname = 'ex%d-%s-me%d_rs%d_nmi%d_nma%d_a%.3f.p' \
%(ex, prob_label, n_methods, reps, min(ns), max(ns), alpha,)
glo.ex_save_result(ex, results, fname)
logger.info('Saved aggregated results to %s' % fname)
home = expanduser("~")
foldername = os.sep.join([home, "minimal_example"])
logger.info("Setting engine folder to %s" % foldername)
# create parameter instance that is needed for any batch computation engine
logger.info("Creating batch parameter instance")
batch_parameters = BatchClusterParameters(foldername=foldername)
# possibly create SGE engine instance, which can be used to submit jobs to
# there are more engines available.
# logger.info("creating SGE engine instance")
# engine = SGEComputationEngine(batch_parameters, check_interval=1)
# # create serial engine (which works locally)
logger.info("Creating serial engine instance")
engine = SerialComputationEngine()
# we have to collect aggregators somehow
aggregators = []
# submit job three times
logger.info("Starting loop over job submission")
for i in range(3):
logger.info("Submitting job %d" % i)
job = MyJob(ScalarResultAggregator())
aggregators.append(engine.submit_job(job))
# let the engine finish its business
logger.info("Wait for all call in engine")
engine.wait_for_all()
this script, we can collect results from the cluster and potentially
submit more jobs.
"""
Log.set_loglevel(10)
# oflder for all job files
home = expanduser("~")
foldername = os.sep.join([home, "minimal_example"])
# parameters for the cluster (folder, name, etcI
batch_parameters = BatchClusterParameters(foldername=foldername)
# engine is the objects that jobs are submitted to
# there are implementations for different batch cluster systems
# the serial one runs everything locally
engine = SerialComputationEngine()
# engine = SGEComputationEngine(batch_parameters)
# engine = SlurmComputationEngine(batch_parameters)
# On submission, the engine returns aggregators that can be
# used to retreive results after potentially doing postprocessing
returned_aggregators = []
for i in range(3):
job = MyJob(ScalarResultAggregator())
agg = engine.submit_job(job)
returned_aggregators.append(agg)
# This call blocks until all jobs are finished (magic happens here)
logger.info("Waiting for all jobs to be completed.")
engine.wait_for_all()
def compute(fname_base,
job_generator,
Ds,
Ns,
num_repetitions,
num_steps,
step_size,
max_steps=None,
compute_local=False):
if not FileSystem.cmd_exists("sbatch") or compute_local:
engine = SerialComputationEngine()
else:
johns_slurm_hack = "#SBATCH --partition=intel-ivy,wrkstn,compute"
folder = os.sep + os.sep.join(["nfs", "data3", "ucabhst", fname_base])
batch_parameters = BatchClusterParameters(
foldername=folder,
resubmit_on_timeout=False,
parameter_prefix=johns_slurm_hack)
engine = SlurmComputationEngine(batch_parameters,
check_interval=1,
do_clean_up=True)
engine.max_jobs_in_queue = 1000
engine.store_fire_and_forget = True
# fixed order of aggregators
aggregators = []
for D in Ds:
for N in Ns:
for j in range(num_repetitions):
logger.info("%s trajectory, D=%d/%d, N=%d/%d repetition %d/%d" % \
(str(job_generator), D, np.max(Ds), N, np.max(Ns), j + 1, num_repetitions))
job = job_generator(D, N, N)
aggregators += [engine.submit_job(job)]
time.sleep(0.1)
# block until all done
engine.wait_for_all()
avg_accept = np.zeros((num_repetitions, len(Ds), len(Ns)))
avg_accept_est = np.zeros((num_repetitions, len(Ds), len(Ns)))
log_dets = np.zeros((num_repetitions, len(Ds), len(Ns)))
log_dets_est = np.zeros((num_repetitions, len(Ds), len(Ns)))
avg_steps_taken = np.zeros((num_repetitions, len(Ds), len(Ns)))
agg_counter = 0
for i in range(len(Ds)):
for k in range(len(Ns)):
for j in range(num_repetitions):
agg = aggregators[agg_counter]
agg_counter += 1
agg.finalize()
result = agg.get_final_result()
agg.clean_up()
avg_accept[j, i, k] = result.acc_mean
avg_accept_est[j, i, k] = result.acc_est_mean
log_dets[j, i, k] = result.vol
log_dets_est[j, i, k] = result.vol_est
avg_steps_taken[j, i, k] = result.steps_taken
with open(fname_base + ".csv", 'a+') as f:
line = np.array([
Ds[i],
Ns[k],
avg_accept[j, i, k],
avg_accept_est[j, i, k],
log_dets[j, i, k],
log_dets_est[j, i, k],
avg_steps_taken[j, i, k],
])
f.write(" ".join(map(str, line)) + os.linesep)