def load_configuration(neonate, verbose=False):
current_file = Path(os.path.abspath(''))
config_file = os.path.join(current_file.parents[1], 'config_files', 'abc',
'neo_config.json')
with open(config_file, 'r') as conf_f:
conf = json.load(conf_f)
params = conf['priors']
input_path = os.path.join(current_file.parents[1], 'data',
'formatted_data',
'{}_formatted.csv'.format(neonate))
d0 = import_actual_data(input_path)
targets = conf['targets']
model_name = conf['model_name']
inputs = conf['inputs']
config = {
"model_name": model_name,
"targets": targets,
"times": d0['t'],
"inputs": inputs,
"parameters": params,
"input_path": input_path,
"zero_flag": conf['zero_flag'],
}
if verbose:
pprint(config)
return config, d0
def load_configuration(model_version, dataset, verbose=False):
current_file = Path(os.path.abspath(''))
config_file = os.path.join(
current_file.parents[2], 'config_files', 'abc',
'bp_hypothermia_{}'.format(model_version),
'bp_hypothermia_{}_config.json'.format(model_version))
with open(config_file, 'r') as conf_f:
conf = json.load(conf_f)
params = conf['priors']
input_path = os.path.join(
current_file.parents[2], 'data', 'clean_hypothermia',
'{}_filtered_formatted.csv'.format(dataset.upper()))
d0 = import_actual_data(input_path)
targets = conf['targets']
model_name = conf['model_name']
inputs = conf['inputs']
config = {
"model_name": model_name,
"targets": targets,
"inputs": inputs,
"parameters": params,
"input_path": input_path,
"zero_flag": conf['zero_flag'],
}
if verbose:
pprint(config)
return config, d0
def get_runs(posterior, conf, n_repeats=50):
rand_selection = random.sample(range(posterior.shape[0]), n_repeats)
outputs_list = []
p_names = list(conf['parameters'].keys())
posteriors = posterior[p_names].values
d0 = import_actual_data(conf['input_path'])
input_data = inputParse(d0, conf['inputs'])
while len(outputs_list) < n_repeats:
idx = rand_selection.pop()
print("\tSample {}, idx:{}".format(len(outputs_list), idx))
p = dict(zip(p_names, posteriors[idx]))
_, output = get_output(conf['model_name'],
p,
conf['times'],
input_data,
d0,
conf['targets'],
distance="NRMSE",
zero_flag=conf['zero_flag'])
outputs_list.append(output)
return outputs_list
def get_repeated_outputs(df,
model_name,
parameters,
input_path,
inputs,
targets,
n_repeats,
zero_flag,
neonate,
tolerance=None,
limit=None,
frac=None,
openopt_path=None,
offset=None,
distance='euclidean'):
"""Generate model output and distances multiple times.
Parameters
----------
model_name : :obj:`str`
Names of model. Should match the modeldef file for model being generated
i.e. model_name of 'model`' should have a modeldef file
'model1.modeldef'.
parameters : :obj:`dict` of :obj:`str`: :obj:`tuple`
Dict of model parameters to compare, with value tuple of the prior max
and min.
input_path : :obj:`str`
Path to the true data file
inputs : :obj:`list` of :obj:`str`
List of model inputs.
targets : :obj:`list` of :obj:`str`
List of model outputs against which the model is being optimised.
n_repeats : :obj: `int`
Number of times to generate output data
frac : :obj:`float`
Fraction of results to consider. Should be given as a percentage i.e.
1=1%, 0.1=0.1%
zero_flag : dict
Dictionary of form target(:obj:`str`): bool, where bool indicates
whether to zero that target.
Note: zero_flag keys should match targets list.
openopt_path : :obj:`str` or :obj:`None`
Path to the openopt data file if it exists. Default is None.
offset : :obj:`dict`
Dictionary of offset parameters if they are needed
distance : :obj:`str`, optional
Distance measure. One of 'euclidean', 'manhattan', 'MAE', 'MSE'.
Returns
-------
fig : :obj:`matplotlib.figure`
Figure containing all axes.
"""
p_names = list(parameters.keys())
sorted_df = df.sort_values(by=distance)
if tolerance:
accepted_limit = sum(df[distance].values < tolerance)
elif limit:
accepted_limit = limit
elif frac:
accepted_limit = frac_calculator(sorted_df, frac)
else:
raise ValueError('No limit or fraction given.')
df_list = []
if n_repeats > accepted_limit:
print("Setting number of repeats to quarter of the posterior size\n",
file=sys.stderr)
n_repeats = int(accepted_limit / 4)
d0 = import_actual_data(input_path)
input_data = inputParse(d0, inputs)
true_data = pd.read_csv(input_path)
times = true_data['t'].values
if openopt_path:
openopt_data = pd.read_csv(openopt_path)
if n_repeats > accepted_limit:
raise ValueError(
"Number of requested model runs greater than posterior size:"
"\n\tPosterior Size: {}\n\tNumber of runs: {}".format(
accepted_limit, n_repeats))
rand_selection = list(range(accepted_limit))
random.shuffle(rand_selection)
outputs_list = []
posteriors = sorted_df.iloc[:accepted_limit][p_names].values
select_idx = 0
with Timer("Running repeat outputs"):
for i in range(n_repeats):
try:
idx = rand_selection.pop()
p = dict(zip(p_names, posteriors[idx]))
if offset:
p = {**p, **offset}
output = get_output(model_name,
p,
times,
input_data,
d0,
targets,
distance=distance,
zero_flag=zero_flag)
outputs_list.append(output)
print("Sample {}, idx:{}".format(len(outputs_list), idx))
except (TimeoutError, TimeoutExpired) as e:
print("Timed out for Sample {}, idx:{}".format(
len(outputs_list), idx))
pprint.pprint(p)
rand_selection.insert(0, idx)
except (CalledProcessError) as e:
print("CalledProcessError for Sample {}, idx:{}".format(
len(outputs_list), idx))
pprint.pprint(p)
rand_selection.insert(0, idx)
print("Final number of runs is: {}".format(len(outputs_list)))
d = {"Errors": {}, "Outputs": {}}
d['Errors']['Average'] = np.nanmean([o[0]['TOTAL'] for o in outputs_list])
for target in targets:
d['Errors'][target] = np.nanmean([o[0][target] for o in outputs_list])
d['Outputs'][target] = [o[1][target] for o in outputs_list]
for ii, target in enumerate(targets):
x = [j for j in times for n in range(len(d['Outputs'][target]))]
with Timer('Transposing {}'.format(target)):
y = np.array(d['Outputs'][target]).transpose()
y = y.ravel()
with Timer("Crafting DataFrame for {}".format(target)):
model_name_col = [neonate] * len(x)
target_col = [target] * len(x)
df1 = pd.DataFrame({
"Time": x,
"Posterior": y,
"Neonate": model_name_col,
"Output": target_col
})
with Timer("Appending dataframe for {}".format(target)):
df_list.append(df1.copy())
del df1
return pd.concat(df_list), true_data
current_file.parents[3],
'data', 'ABC', 'nrmse_SA',
MODEL_VERSION,
DATASET)
pfile = os.path.abspath(os.path.join(
output_dir,
'reduced_sorted_parameters.csv')
)
input_path = os.path.join(current_file.parents[3],
'data',
'clean_hypothermia',
'{}_filtered_formatted.csv'.format(DATASET.upper()))
d0 = import_actual_data(input_path)
targets = conf['targets']
model_name = conf['model_name']
inputs = conf['inputs']
config = {
"model_name": model_name,
"targets": targets,
"inputs": inputs,
"parameters": params,
"input_path": input_path,
"zero_flag": conf['zero_flag']
}
pprint(config)
