def model_diff(parameter_set_nr, data_kind='WOA', path='/tmp', normalize_with_deviation=False, y_max=(None, None)):
from simulation.model.constants import DATABASE_OUTPUT_DIR, DATABASE_MODEL_DIRNAME, DATABASE_TIME_STEP_DIRNAME, DATABASE_PARAMETERS_SET_DIRNAME, DATABASE_PARAMETERS_FILENAME
from simulation.model.constants import (METOS_X_DIM as X_DIM, METOS_Y_DIM as Y_DIM, METOS_Z_LEFT as Z_VALUES_LEFT)
logger.debug('Plotting model output for parameter set {}'.format(parameter_set_nr))
## load parameters
parameter_set_dirname = DATABASE_PARAMETERS_SET_DIRNAME.format(parameter_set_nr)
p_file = os.path.join(DATABASE_OUTPUT_DIR, DATABASE_MODEL_DIRNAME.format('dop_po4'), DATABASE_TIME_STEP_DIRNAME.format(1), parameter_set_dirname, DATABASE_PARAMETERS_FILENAME)
p = np.loadtxt(p_file)
## init data base
data_base = simulation.util.data_base.init_data_base(data_kind)
if not normalize_with_deviation:
file = os.path.join(path, 'model_diff_-_{}_-_' + parameter_set_dirname + '_-_{}.png')
else:
file = os.path.join(path, 'model_diff_normalized_with_deviation_-_{}_-_' + parameter_set_dirname + '_-_{}.png')
## print for WOA
if data_kind.upper() == 'WOA':
diff_boxes = np.abs(data_base.diff_boxes(p, normalize_with_deviation=normalize_with_deviation))
## print for WOD
elif data_kind.upper() == 'WOD':
diff = np.abs(data_base.diff(p, normalize_with_deviation=normalize_with_deviation))
diff_boxes = data_base.convert_to_boxes(diff, no_data_value=np.inf)
else:
raise ValueError('Data_kind {} unknown. Must be "WOA" or "WOD".'.format(data_kind))
def plot_tracer_diff(diff, file, y_max=None):
util.plot.data(np.abs(diff), file, land_value=np.nan, no_data_value=np.inf, v_min=0, v_max=y_max)
plot_tracer_diff(diff_boxes[0], file.format(data_kind, 'dop'), y_max=y_max[0])
plot_tracer_diff(diff_boxes[1], file.format(data_kind, 'po4'), y_max=y_max[1])
def model_confidence(parameter_set_nr, kind='WOA_WLS', path='/tmp', v_max=[None, None], time_dim_confidence=12, time_dim_df=12, average_in_time=False):
from simulation.model.constants import DATABASE_OUTPUT_DIR, DATABASE_MODEL_DIRNAME, DATABASE_TIME_STEP_DIRNAME, DATABASE_PARAMETERS_SET_DIRNAME
from simulation.accuracy.constants import CACHE_DIRNAME, MODEL_CONFIDENCE_FILENAME
logger.debug('Plotting model confidence for parameter set {}'.format(parameter_set_nr))
## load value
parameter_set_dirname = DATABASE_PARAMETERS_SET_DIRNAME.format(parameter_set_nr)
f_file = os.path.join(DATABASE_OUTPUT_DIR, DATABASE_MODEL_DIRNAME.format('dop_po4'), DATABASE_TIME_STEP_DIRNAME.format(1), parameter_set_dirname, CACHE_DIRNAME, kind, MODEL_CONFIDENCE_FILENAME.format(time_dim_confidence=time_dim_confidence, time_dim_df=time_dim_df))
f = np.load(f_file)
if average_in_time:
f = f.mean(axis=1)
assert len(f) == 2
## set v_max
for i in range(len(f)):
if v_max[i] is None:
v_max[i] = np.nanmax(f[i])
rounding_exponent = np.sign(np.log10(v_max[i])) * np.ceil(np.abs(np.log10(v_max[i])))
v_max[i] = np.floor(v_max[i] * 10**(-rounding_exponent)) * 10**rounding_exponent
## plot
file = os.path.join(path, 'model_confidence_-_' + parameter_set_dirname + '_-_' + get_label(kind) + '_-_time_dim_df_{}'.format(time_dim_df) + '_-_{tracer}.png')
tracers = ('dop', 'po4')
for i in range(len(tracers)):
util.plot.data(f[i], file.format(tracer=tracers[i]), land_value=np.nan, no_data_value=None, v_min=0, v_max=v_max[i], contours=True, colorbar=False)
def model_output(parameter_set_nr, kind='BOXES', path='/tmp', y_max=(None, None), average_in_time=False):
from simulation.model.constants import DATABASE_OUTPUT_DIR, DATABASE_MODEL_DIRNAME, DATABASE_TIME_STEP_DIRNAME, DATABASE_PARAMETERS_SET_DIRNAME, DATABASE_PARAMETERS_FILENAME
from simulation.util.constants import CACHE_DIRNAME, BOXES_F_FILENAME, WOD_F_FILENAME
logger.debug('Plotting model output for parameter set {}'.format(parameter_set_nr))
## load parameters
parameter_set_dirname = DATABASE_PARAMETERS_SET_DIRNAME.format(parameter_set_nr)
p_file = os.path.join(DATABASE_OUTPUT_DIR, DATABASE_MODEL_DIRNAME.format('dop_po4'), DATABASE_TIME_STEP_DIRNAME.format(1), parameter_set_dirname, DATABASE_PARAMETERS_FILENAME)
p = np.loadtxt(p_file)
## init data base
if kind.upper() == 'BOXES':
data_base = simulation.util.data_base.init_data_base('WOA')
f = data_base.f_boxes(p)
f[f < 0] = 0
if average_in_time:
f = f.mean(axis=1)
else:
data_base = simulation.util.data_base.init_data_base('WOD')
f = data_base.f(p)
f = data_base.convert_to_boxes(f, no_data_value=np.inf)
file = os.path.join(path, 'model_output_-_' + kind + '_-_' + parameter_set_dirname + '_-_{tracer}.png')
tracers = ('dop', 'po4')
for i in range(len(tracers)):
util.plot.data(f[i], file.format(tracer=tracers[i]), land_value=np.nan, no_data_value=np.inf, v_min=0, v_max=y_max[i], contours=True, colorbar=False)
def relative_parameter_confidence(parameter_set_nr, kind='WOA_WLS', path='/tmp'):
from simulation.model.constants import DATABASE_OUTPUT_DIR, DATABASE_MODEL_DIRNAME, DATABASE_TIME_STEP_DIRNAME, DATABASE_PARAMETERS_SET_DIRNAME, DATABASE_PARAMETERS_FILENAME
from simulation.accuracy.constants import CACHE_DIRNAME, PARAMETER_CONFIDENCE_FILENAME
logger.debug('Plotting parameter confidence for parameter set {}'.format(parameter_set_nr))
## load value
parameter_dirname = DATABASE_PARAMETERS_SET_DIRNAME.format(parameter_set_nr)
parameter_dir = os.path.join(DATABASE_OUTPUT_DIR, DATABASE_MODEL_DIRNAME.format('dop_po4'), DATABASE_TIME_STEP_DIRNAME.format(1), parameter_dirname)
parameter_value_file = os.path.join(parameter_dir, DATABASE_PARAMETERS_FILENAME)
parameter_value = np.loadtxt(parameter_value_file)
parameter_confidence_file = os.path.join(parameter_dir, CACHE_DIRNAME, kind, PARAMETER_CONFIDENCE_FILENAME)
parameter_confidence = np.load(parameter_confidence_file)
relative_parameter_confidence_percent = 100 * parameter_confidence / parameter_value
relative_parameter_confidence_percent_interval = [relative_parameter_confidence_percent, - relative_parameter_confidence_percent]
## plot
file = os.path.join(path, 'relative_parameter_confidence_-_' + parameter_dirname + '_-_' + get_label(kind) + '.png')
util.plot.intervals(relative_parameter_confidence_percent_interval, file, use_percent_ticks=True)
def average_model_confidence_increase(parameter_set_nr, kind='WOA_WLS', path='/tmp', time_dim_df=12):
from simulation.model.constants import DATABASE_OUTPUT_DIR, DATABASE_MODEL_DIRNAME, DATABASE_TIME_STEP_DIRNAME, DATABASE_PARAMETERS_SET_DIRNAME
from simulation.accuracy.constants import CACHE_DIRNAME, AVERAGE_MODEL_CONFIDENCE_INCREASE_FILENAME
logger.debug('Plotting average model confidence increase for parameter set {}'.format(parameter_set_nr))
parameter_set_dirname = DATABASE_PARAMETERS_SET_DIRNAME.format(parameter_set_nr)
f_file = os.path.join(DATABASE_OUTPUT_DIR, DATABASE_MODEL_DIRNAME.format('dop_po4'), DATABASE_TIME_STEP_DIRNAME.format(1), parameter_set_dirname, CACHE_DIRNAME, kind, AVERAGE_MODEL_CONFIDENCE_INCREASE_FILENAME.format(time_dim_df))
f = np.load(f_file)
v_min = np.nanmin(f, axis=tuple(np.arange(f.ndim-1)+1))
v_max = np.nanmax(f, axis=tuple(np.arange(f.ndim-1)+1))
significant_digits=3
for i in range(len(v_min)):
round_factor = 10 ** (np.ceil(-np.log10(v_min[i])) + significant_digits)
v_min[i] = np.floor(v_min[i] * round_factor) / round_factor
for i in range(len(v_max)):
round_factor = 10 ** (np.ceil(-np.log10(v_min[i])) + significant_digits)
v_max[i] = np.ceil(v_max[i] * round_factor) / round_factor
file = os.path.join(path, 'average_model_confidence_increase_-_' + parameter_set_dirname + '_-_time_dim_df_{}'.format(time_dim_df) + '_-_{}_-_{}.png')
util.plot.data(f[0], file.format(kind, 'dop'), land_value=np.nan, no_data_value=None, v_min=v_min[0], v_max=v_max[0])
util.plot.data(f[1], file.format(kind, 'po4'), land_value=np.nan, no_data_value=None, v_min=v_min[1], v_max=v_max[1])
def save(model_name='dop_po4', time_step=1, parameter_sets=range(9999), data_kind='WOA', eval_f=True, eval_df=True, as_jobs=False, node_kind='clexpress'):
from simulation.model.constants import DATABASE_OUTPUT_DIR, DATABASE_MODEL_DIRNAME, DATABASE_TIME_STEP_DIRNAME, DATABASE_PARAMETERS_SET_DIRNAME, DATABASE_PARAMETERS_FILENAME, DATABASE_SPINUP_DIRNAME, DATABASE_DERIVATIVE_DIRNAME
## get time step dir
model_dirname = DATABASE_MODEL_DIRNAME.format(model_name)
model_dir = os.path.join(DATABASE_OUTPUT_DIR, model_dirname)
time_step_dirname = DATABASE_TIME_STEP_DIRNAME.format(time_step)
time_step_dir = os.path.join(model_dir, time_step_dirname)
## create model
model = simulation.model.eval.Model()
## save for all parameter sets
for parameter_set_number in parameter_sets:
## get parameter
cost_function_family = None
parameter_set_dirname = DATABASE_PARAMETERS_SET_DIRNAME.format(parameter_set_number)
parameter_set_dir = os.path.join(time_step_dir, parameter_set_dirname)
parameters_file = os.path.join(parameter_set_dir, DATABASE_PARAMETERS_FILENAME)
## create cost functions
if os.path.exists(parameters_file):
spinup_dir = os.path.join(parameter_set_dir, DATABASE_SPINUP_DIRNAME)
last_run_dir = model.last_run_dir(spinup_dir)
if last_run_dir is not None:
cf_kargs = {'data_kind': data_kind, 'model_options': {'spinup_options': {'years':1,}, 'time_step': time_step, 'total_concentration_factor_included_in_parameters': True}, 'job_setup':{'name': 'SCF_' + data_kind}}
cost_function_family = simulation.optimization.cost_function.Family(**cf_kargs)
## eval cf family
if cost_function_family is not None:
p = np.loadtxt(parameters_file)
try:
## eval cf by itself
if not as_jobs:
## eval f
if eval_f:
cost_function_family.f(p)
cost_function_family.f_normalized(p)
## eval df
if eval_df:
derivative_dir = os.path.join(parameter_set_dir, DATABASE_DERIVATIVE_DIRNAME)
if os.path.exists(derivative_dir):
cost_function_family.df(p)
## eval cf as job
else:
for cf in cost_function_family.family:
if (eval_f and not cf.f_available(p)) or (eval_df and not cf.df_available(p)):
from util.constants import TMP_DIR
output_dir = tempfile.TemporaryDirectory(dir=TMP_DIR, prefix='save_value_cost_function_tmp_').name
cf_kargs = cf.kargs
cf_kargs['job_setup'] = {'name': '{}:{}'.format(cf, parameter_set_number)}
nodes_setup = util.batch.universal.system.NodeSetup(memory=50, node_kind=node_kind, nodes=1, cpus=1, total_cpus_max=1, walltime=1)
with simulation.optimization.job.CostFunctionJob(output_dir, p, cf.kind, eval_f=eval_f, eval_df=eval_df, nodes_setup=nodes_setup, **cf_kargs) as cf_job:
cf_job.start()
time.sleep(10)
except ValueError as e:
logger.exception(e)
