def _parallel_mc(processes, path_stem, calc_formula, l, M, build_output, variable_name, sigma_function, interp_fn, angles_fn, Lmin, Lmax, wavetype, dictionary_type, iteration, normalisation_factor):
"""
Split the tasks so the algorithm be parallelised and then collect the parallel output
"""
# put in queue
in_queue = mp.Queue()
out_queue = mp.Queue()
future_res = []
for i in range(processes):
path = path_stem + str(i) + '/'
if not os.path.exists(path):
os.makedirs(path)
future_res.append(mp.Process(target = calc_formula, args = (path, in_queue, out_queue)))
future_res[-1].start()
for j in range(iteration):
in_queue.put((l, M, build_output, sigma_function, variable_name, Lmin, Lmax, wavetype, interp_fn, angles_fn, normalisation_factor))
# send stop signals
for i in range(processes):
in_queue.put('stop')
# collect output
results = []
for i in range(iteration):
if (i+1)%1000 == 0:
print(i)
mlmc.write(logfile, "%f" % (i))
mlmc.write(logfile, "\n")
results.append(out_queue.get())
outputf = [f[0] for f in results]
outputc = [f[1] for f in results]
outputfc = 0
outputfc2 = 0
outputfc3 = 0
outputfc4 = 0
outputfsum = 0
outputf2sum = 0
# record output values in csv (necessary for doing inverse sampling)
try:
print('output' + str(l+1)+'.csv')
outputprevf = pd.read_csv('output' + str(l+1)+'.csv')
outputf_df = pd.DataFrame(outputf, columns = ['output'])
pd.concat([outputf_df, outputprevf]).to_csv('output' + str(l+1)+'.csv', index = False)
print(len(pd.concat([outputf_df, outputprevf])))
except:
pd.DataFrame(outputf, columns = ['output']).to_csv('output' + str(l+1) + '.csv', index = False)
try:
print('output' + str(l)+'.csv')
outputprevc = pd.read_csv('output' + str(l)+'.csv')
outputc_df = pd.DataFrame(outputc, columns = ['output'])
pd.concat([outputc_df, outputprevc]).to_csv('output' + str(l)+'.csv', index = False)
print(len(pd.concat([outputc_df, outputprevc])))
except:
pd.DataFrame(outputc, columns = ['output']).to_csv('output' + str(l) + '.csv', index = False)
for i in range(len(outputf)):
if np.isnan(outputf[i]):
print("nan")
elif np.isnan(outputc[i]):
print("nan")
else:
outputfc += (outputf[i] - outputc[i])
outputfc2 += ((outputf[i] - outputc[i])**2)
outputfc3 += ((outputf[i] - outputc[i])**3)
outputfc4 += ((outputf[i] - outputc[i])**4)
outputfsum += outputf[i]
outputf2sum += outputf[i]**2
sums = np.array([outputfc, outputfc2, outputfc3, outputfc4, outputfsum, outputf2sum])
return sums
def _parallel_mc(processes, path_stem, calc_formula, l, M, build_output, variable_name, sigma_function, interp_fn, angles_fn, Lmin, Lmax, wavetype, dictionary_type, iteration, normalisation_factor):
"""
Split the tasks so the algorithm be parallelised and then collect the parallel output
"""
# putting runs into queue
in_queue = mp.Queue()
out_queue = mp.Queue()
future_res = []
for i in range(processes):
path = path_stem + str(i) + '/'
if not os.path.exists(path):
os.makedirs(path)
future_res.append(mp.Process(target = calc_formula, args = (path, in_queue, out_queue)))
future_res[-1].start()
for j in range(iteration):
in_queue.put((l, M, build_output, sigma_function, variable_name, Lmin, Lmax, wavetype, interp_fn, angles_fn))
# send stop signals
for i in range(processes):
in_queue.put('stop')
# collect output
results = []
for i in range(iteration):
if (i+1)%1000 == 0:
print(i)
mlmc.write(logfile, "%f" % (i))
mlmc.write(logfile, "\n")
results.append(out_queue.get())
outputf = [f[0] for f in results]
outputc = [f[1] for f in results]
outputfc = 0
outputfc2 = 0
outputfc3 = 0
outputfc4 = 0
outputfsum = 0
outputf2sum = 0
# record output to csv file
pd.DataFrame(outputf).to_csv(filename[:-4] + '.csv')
for i in range(len(outputf)):
if np.isnan(outputf[i]):
print("nan")
elif np.isnan(outputc[i]):
print("nan")
else:
outputfc += (outputf[i] - outputc[i])
outputfc2 += ((outputf[i] - outputc[i])**2)
outputfc3 += ((outputf[i] - outputc[i])**3)
outputfc4 += ((outputf[i] - outputc[i])**4)
outputfsum += outputf[i]
outputf2sum += outputf[i]**2
sums = np.array([outputfc, outputfc2, outputfc3, outputfc4, outputfsum, outputf2sum])
return sums
dellist = []
varlist = []
ts = time.time()
st = datetime.datetime.fromtimestamp(ts).strftime('%Y-%m-%d %H:%M:%S')
# set up a unique file to store outputs
filename = "mlmc_xbeach_" + st + ".txt"
logfile = open(filename, "w+")
logfile.close()
# run MLMC algorithm
for i in range(no_runs):
t0 = time.time()
logfile = open(filename, "a+")#, 0)
mlmc.write(logfile, str(no_parallel_processes))
if init_test == True:
alpha_mod, beta, gamma, cost, var1, var2, del1 = mlmc.mlmc_test(no_parallel_processes, path_stem, _parallel_mc, multilevel, wavetype, dictionary_type, M, L, N0, Lmin, Lmax, logfile, build_output, variable_name, sample, interp_fn = interp_fn, normalisation_factor = init_vol)
vardifflist.append(var1)
varsing.append(var2)
costlist.append(cost)
dellist.append(del1)
expectedlist.append(sum(del1))
if eps_test == True:
# these values come from the results of init_test and thus must be replaced if a new init_test run is performed
alpha_mod = 0.921053
beta = 1.970078
gamma = 2.400694
var1 = [5.9081e-07, 6.3646e-08, 1.9078e-08, 1.0362e-08, 1.5862e-09]
del1 = [1.0026e-03, 1.2740e-04, 1.0285e-4, 9.3601e-05, 4.8057e-05]
Ns, varlevel, Plist, P_seq_list, cost_per_epsilon = mlmc.eps_only(alpha_mod, beta, gamma, var1, del1, no_parallel_processes, path_stem, _parallel_mc, multilevel, wavetype, dictionary_type, M, L, 2, epsilon, Lmin, Lmax, logfile, build_output, variable_name, sample, interp_fn = interp_fn, angles_fn = 0, normalisation_factor = init_vol)
variable_name = 'Hm0' # name of uncertain variable
dictionary_type = 'w' # dictionary where uncertain variable lives
costlist = []
dellist = []
# set output txt file
ts = time.time()
st = datetime.datetime.fromtimestamp(ts).strftime('%Y-%m-%d %H:%M:%S')
filename = "mc_xbeach_" + st + ".txt"
logfile = open(filename, "w+")
logfile.close()
t0 = time.time()
logfile = open(filename, "a+")#, 0)
mlmc.write(logfile, str(no_parallel_processes))
# run monte carlo algorithm
cost, del1 = mlmc.mlmc_test(no_parallel_processes, path_stem, _parallel_mc, multilevel, 'jonstable', dictionary_type, M, L, N0, Lmin, Lmax, logfile, build_output, variable_name, sample, interp_fn = interp_fn, angles_fn = angle_dict)
costlist.append(cost)
dellist.append(del1)
# record cost and expectation from this one run
logfile = open(filename, "a+")#, 0)
print('total time: ' + str(t1 - t0))
mlmc.write(logfile, "total time: %f" % (t1-t0))
mlmc.write(logfile, "\n")