mirror of
https://github.com/ANL-CEEESA/LLEPE.git
synced 2025-12-06 09:58:52 -06:00
added example file to do iterative optimization
This commit is contained in:
titusquah
194
docs/Examples/iterative_fitter.py
Normal file
194
docs/Examples/iterative_fitter.py
Normal file
@@ -0,0 +1,194 @@
|
||||
from scipy.optimize import curve_fit
|
||||
import llepe
|
||||
import pandas as pd
|
||||
import numpy as np
|
||||
import json
|
||||
|
||||
|
||||
def linear(x, a, b):
|
||||
return a * x + b
|
||||
|
||||
|
||||
species_list = 'Nd,Pr,Ce,La,Dy,Sm,Y'.split(',')
|
||||
pitzer_param_list = ['beta0', 'beta1', 'Cphi']
|
||||
meas_pitzer_param_df = pd.read_csv("../../data/csvs/may_pitzer_params.csv")
|
||||
labeled_data = pd.read_csv("../../data/csvs/"
|
||||
"zeroes_removed_PC88A_HCL_NdPrCeLaDySmY.csv")
|
||||
exp_data = labeled_data.drop(labeled_data.columns[0], axis=1)
|
||||
xml_file = "PC88A_HCL_NdPrCeLaDySmY_w_pitzer.xml"
|
||||
eps = 1e-4
|
||||
mini_eps = 1e-8
|
||||
x_guesses = [[-5178500.0, -1459500.0],
|
||||
[-5178342.857142857, -1460300.0],
|
||||
[-5178342.857142857, -1459500.0],
|
||||
[-5178342.857142857, -1458300.0],
|
||||
[-5178185.714285715, -1459900.0],
|
||||
[-5178185.714285715, -1459500.0],
|
||||
[-5178185.714285715, -1459100.0],
|
||||
[-5178185.714285715, -1458300.0],
|
||||
[-5178028.571428572, -1459900.0],
|
||||
[-5178028.571428572, -1459100.0],
|
||||
[-5178028.571428572, -1458300.0],
|
||||
[-5177557.142857143, -1459900.0],
|
||||
[-5177400.0, -1460300.0]]
|
||||
pitzer_guess_df = meas_pitzer_param_df.copy()
|
||||
ignore_list = []
|
||||
optimizer = 'scipy_minimize'
|
||||
output_dict = {'iter': [0],
|
||||
'best_obj': [1e20],
|
||||
'rel_diff': [1e20]}
|
||||
for species in species_list:
|
||||
output_dict['{0}_slope'.format(species)] = [1e20]
|
||||
output_dict['{0}_intercept'.format(species)] = [1e20]
|
||||
for pitzer_param in pitzer_param_list:
|
||||
output_dict['{0}_{1}'.format(species, pitzer_param)] = [1e20]
|
||||
i = 0
|
||||
rel_diff = 1000
|
||||
while rel_diff > 1e-4:
|
||||
i += 1
|
||||
best_obj = 1e20
|
||||
output_dict['iter'].append(i)
|
||||
for species in species_list:
|
||||
lower_species = species.lower()
|
||||
opt_values = {
|
||||
'(HA)2(org)_h0': [],
|
||||
'{0}(H(A)2)3(org)_h0'.format(species): [],
|
||||
'beta0': [],
|
||||
'beta1': [],
|
||||
'Cphi': [],
|
||||
'obj_value': [],
|
||||
'guess': []}
|
||||
for x_guess in x_guesses:
|
||||
info_dict = {'(HA)2(org)_h0': {'upper_element_name': 'species',
|
||||
'upper_attrib_name': 'name',
|
||||
'upper_attrib_value': '(HA)2(org)',
|
||||
'lower_element_name': 'h0',
|
||||
'lower_attrib_name': None,
|
||||
'lower_attrib_value': None,
|
||||
'input_format': '{0}',
|
||||
'input_value': x_guess[1]},
|
||||
'{0}(H(A)2)3(org)_h0'.format(species): {
|
||||
'upper_element_name': 'species',
|
||||
'upper_attrib_name': 'name',
|
||||
'upper_attrib_value': '{0}(H(A)2)3(org)'.format(
|
||||
species),
|
||||
'lower_element_name': 'h0',
|
||||
'lower_attrib_name': None,
|
||||
'lower_attrib_value': None,
|
||||
'input_format': '{0}',
|
||||
'input_value': x_guess[0]},
|
||||
}
|
||||
for pitzer_param in pitzer_param_list:
|
||||
if '{0}_{1}'.format(species, pitzer_param) not in ignore_list:
|
||||
pitzer_row = pitzer_guess_df[
|
||||
pitzer_guess_df['species'] == species]
|
||||
inner_dict = {'upper_element_name': 'binarySaltParameters',
|
||||
'upper_attrib_name': 'cation',
|
||||
'upper_attrib_value':
|
||||
'{0}+++'.format(species),
|
||||
'lower_element_name': pitzer_param,
|
||||
'lower_attrib_name': None,
|
||||
'lower_attrib_value': None,
|
||||
'input_format': ' {0}, 0.0, 0.0, 0.0, 0.0 ',
|
||||
'input_value':
|
||||
pitzer_row[pitzer_param].values[0]
|
||||
}
|
||||
info_dict['{0}_{1}'.format(
|
||||
species, pitzer_param)] = inner_dict
|
||||
llepe_params = {
|
||||
'exp_data': exp_data,
|
||||
'phases_xml_filename': xml_file,
|
||||
'opt_dict': info_dict,
|
||||
'phase_names': ['HCl_electrolyte', 'PC88A_liquid'],
|
||||
'aq_solvent_name': 'H2O(L)',
|
||||
'extractant_name': '(HA)2(org)',
|
||||
'diluant_name': 'dodecane',
|
||||
'complex_names': ['{0}(H(A)2)3(org)'.format(species)],
|
||||
'extracted_species_ion_names': ['{0}+++'.format(species)],
|
||||
'aq_solvent_rho': 1000.0,
|
||||
'extractant_rho': 960.0,
|
||||
'diluant_rho': 750.0,
|
||||
'objective_function': llepe.lmse_perturbed_obj,
|
||||
'optimizer': optimizer
|
||||
}
|
||||
estimator = llepe.LLEPE(**llepe_params)
|
||||
estimator.update_xml(llepe_params['opt_dict'])
|
||||
obj_kwargs = {'species_list': [species], 'epsilon': 1e-100}
|
||||
bounds = [(1e-1, 1e1)] * len(info_dict)
|
||||
optimizer_kwargs = {"method": 'l-bfgs-b',
|
||||
"bounds": bounds}
|
||||
opt_dict, obj_value = estimator.fit(
|
||||
objective_kwargs=obj_kwargs,
|
||||
optimizer_kwargs=optimizer_kwargs)
|
||||
if obj_value < best_obj:
|
||||
best_obj = obj_value
|
||||
keys = list(opt_dict.keys())
|
||||
info1 = [opt_dict[key]['input_value'] for key in keys]
|
||||
info1.append(obj_value)
|
||||
info1.append(x_guess)
|
||||
opt_values_keys = opt_values.keys()
|
||||
for ind, key in enumerate(opt_values_keys):
|
||||
opt_values[key].append(info1[ind])
|
||||
opt_value_df = pd.DataFrame(opt_values)
|
||||
p_opt, p_cov = curve_fit(linear,
|
||||
opt_value_df['(HA)2(org)_h0'].values,
|
||||
opt_value_df['{0}(H(A)2)3(org)_h0'.format(
|
||||
species)].values)
|
||||
slope, intercept = p_opt
|
||||
output_dict['{0}_slope'.format(species)].append(slope)
|
||||
output_dict['{0}_intercept'.format(species)].append(intercept)
|
||||
min_h0_df = opt_value_df[
|
||||
opt_value_df['(HA)2(org)_h0']
|
||||
== opt_value_df['(HA)2(org)_h0'].min()]
|
||||
update_pitzer_dict = {}
|
||||
for pitzer_param in pitzer_param_list:
|
||||
key_name = '{0}_{1}'.format(species, pitzer_param)
|
||||
output_dict[key_name].append(min_h0_df[pitzer_param].values[0])
|
||||
|
||||
inner_dict = {'upper_element_name': 'binarySaltParameters',
|
||||
'upper_attrib_name': 'cation',
|
||||
'upper_attrib_value':
|
||||
'{0}+++'.format(species),
|
||||
'lower_element_name': pitzer_param,
|
||||
'lower_attrib_name': None,
|
||||
'lower_attrib_value': None,
|
||||
'input_format': ' {0}, 0.0, 0.0, 0.0, 0.0 ',
|
||||
'input_value':
|
||||
min_h0_df[pitzer_param].values[0]
|
||||
}
|
||||
update_pitzer_dict['{0}_{1}'.format(
|
||||
species, pitzer_param)] = inner_dict
|
||||
estimator.update_xml(update_pitzer_dict)
|
||||
|
||||
pitzer_guess_dict = {'species': [],
|
||||
'beta0': [],
|
||||
'beta1': [],
|
||||
'Cphi': []}
|
||||
for species in species_list:
|
||||
pitzer_guess_dict['species'].append(species)
|
||||
for pitzer_param in pitzer_param_list:
|
||||
value_list = output_dict['{0}_{1}'.format(species, pitzer_param)]
|
||||
value = value_list[-1]
|
||||
pitzer_guess_dict[pitzer_param].append(value)
|
||||
if i > 2:
|
||||
mini_rel_diff1 = np.abs(value_list[-1]
|
||||
- value_list[-2])/(
|
||||
np.abs(value_list[-2]))
|
||||
mini_rel_diff2 = np.abs(value_list[-2] - value_list[-3]) / (
|
||||
np.abs(value_list[-3]))
|
||||
if mini_rel_diff1 < mini_eps and mini_rel_diff2 < mini_eps:
|
||||
ignore_list.append('{0}_{1}'.format(species, pitzer_param))
|
||||
pitzer_guess_df = pd.DataFrame(pitzer_guess_dict)
|
||||
|
||||
output_dict['best_obj'].append(best_obj)
|
||||
output_df = pd.DataFrame(output_dict)
|
||||
old_row = output_df.iloc[-2, :].values[3:]
|
||||
new_row = output_df.iloc[-1, :].values[3:]
|
||||
rel_diff = np.sum(np.abs(new_row - old_row)/np.abs(old_row))
|
||||
output_dict['rel_diff'].append(rel_diff)
|
||||
output_df = pd.DataFrame(output_dict)
|
||||
output_df.to_csv('outputs/iterative_fitter_output_df.csv')
|
||||
|
||||
|
||||
|
||||
|
||||
Reference in New Issue
Block a user