def __init__(self, history, parameters, basename="out"):
"""Creates an experiment class for uncertainty analysis methods for kinematic models
**Arguments**:
- *history* = The .his file this experiment is based on
- *parameters* = A string pointing to a csv file defining the statistical
properties of the model properties being varied, or alternatively an array
of python dictionaries with the same function. This file/dictionary array
should have collumns/keys defining:
1) the event and parameter being varied (titled 'event' and 'parameter')
2) the statistical distribution to sample from (titled 'type' and containing either 'normal',
'vonmises' or 'uniform')
3) the distribution mean (titled 'mean') and,
4) a collumn defining the distance between the 2.5th and 97.5th percentiles
(titled '+-') OR one defining the standard deviation (titled 'stdev')
"""
# init monte carlo class
MonteCarlo.__init__(self, history, parameters, basename)
# add empty block (otherwise something breaks...)
self.block = None
def estimate_uncertainty(self, n_trials, **kwds):
"""
Samples the specified number of models, given the pdf's defined in the params file used to create this model.
**Arguments**:
- *n_trials* = The number of random draws to produce. The variation between these random draws
is used to estimate uncertainty.
**Optional Keywords**:
- *verbose* = If true, this funciton prints information to the print buffer. Default is True.
- *model_path* = The directory to write models to. Default is a local directory called 'tmp'.
- *cleanup* = True if this function should delete any models it creates (they're not needed anymore). Default
is True.
"""
vb = kwds.get('verbose', False)
model_path = kwds.get('model_path', 'tmp')
cleanup = kwds.get('cleanup', True)
# generate & load initial model
self.write_history('tmp.his')
pynoddy.compute_model('tmp.his', self.basename)
self.load_model_info()
self.load_geology()
os.remove('tmp.his')
# perform monte carlo sampling
if vb:
print "Producing model realisations..."
self.generate_model_instances(model_path, n_trials, verbose=vb, write_changes=None)
# thought: it would be more efficient (memory wise) to load models 1 at a time rather than
# dumping them all in memory....
# load results
if vb:
print "Loading models..."
models = MonteCarlo.load_noddy_realisations(model_path, verbose=vb)
self.models = models
# compute strat column
# self.determine_model_stratigraphy()
# self.n_rocktypes = len(self.model_stratigraphy)
# self.nx = models[0].nx
# self.ny = models[0].ny
# self.nz = models[0].nz
# calculate probabilities for each lithology. p_block[lithology][x][y][z] = p(lithology | x, y ,z)
self.p_block = [[[[0. for z in range(self.nz)] for y in range(self.ny)] for x in range(self.nx)] for l in
range(self.n_rocktypes)]
p1 = 1 / float(n_trials) # probability increment gained on each observation
for m in models:
# loop through voxels
for x in range(self.nx):
for y in range(self.ny):
for z in range(self.nz):
# get litho
litho = int(m.block[x][y][z]) - 1
# update litho probability
self.p_block[litho][x][y][z] += p1
# calculate entropy & store in self.e_block
self.e_block = np.ndarray((self.nx, self.ny, self.nz))
for x in range(self.nx):
for y in range(self.ny):
for z in range(self.nz):
entropy = 0 # calculate shannons information entropy
for litho in range(self.n_rocktypes):
p = self.p_block[litho][x][y][z]
# fix domain to 0 < p < 1
if p == 0:
p = 0.0000000000000001
if p >= 0.9999999999999999:
p = 0.9999999999999999
# calculate
entropy += p * math.log(p, 2) + (1 - p) * (math.log(1 - p, 2))
entropy = entropy * -1 / float(self.n_rocktypes) # divide by n
self.e_block[x][y][z] = entropy
# cleanup
if vb:
print "Cleaning up..."
if cleanup:
self.cleanup()
if vb:
print "Finished."
#generate random purturbation
ex1.random_draw()
except Exception as e:
sys.stderr.write("Error - could not randomly purturb the Experiment... %s\n" % e)
err = True
if not err:
print("Succesfully used Experiment class")
###########################
#Test MonteCarlo class
###########################
try:
mc = MonteCarlo(history_path,params)
mc.generate_model_instances("out",4,threads=4,verbose=False,write_changes=None)
mc.cleanup() #delete files
except Exception as e:
sys.stderr.write("Error - MonteCarlo class is not functioning... %s\n" % e)
err = True
if not err:
print("Succesfully used MonteCarlo class")
#cleanup
os.remove(history_path)
import glob
for filename in glob.glob("%s*" % output_name):
os.remove(filename)
os.remove('noddyBatchProgress.txt')
#generate random purturbation
ex1.random_draw()
except Exception as e:
sys.stderr.write(
"Error - could not randomly purturb the Experiment... %s\n" % e)
err = True
if not err:
print "Succesfully used Experiment class"
###########################
#Test MonteCarlo class
###########################
try:
mc = MonteCarlo(history_path, params)
mc.generate_model_instances("out",
4,
threads=4,
verbose=False,
write_changes=None)
mc.cleanup() #delete files
except Exception as e:
sys.stderr.write("Error - MonteCarlo class is not functioning... %s\n" % e)
err = True
if not err:
print "Succesfully used MonteCarlo class"
#cleanup
os.remove(history_path)
