def LNKS_fobj_helper(f, theta, stim, y, options):
'''
LNKS model objective function helper function
Weighted sum of log-likelihood and mean-square error
'''
model = options['model']
if model == 'LNK':
y_est = f(theta, stim, options)
J = _obj.mse_weighted_loss(y, y_est, len_section=10000, weight_type='std')
elif model == 'LNKS':
y_est = f(theta, stim, options)
# linear combination of objective functions
J_mse = _obj.mse_weighted_loss(y, y_est, len_section=10000, weight_type='mean')
J_poss = _obj.poisson_weighted_loss(y, y_est, len_section=10000, weight_type="mean")
J = J_poss + J_mse
elif model == 'LNKS_MP':
# data
y_mp = y[0]
y_fr = y[1]
y_mp_est, y_fr_est = f(theta, stim, options)
# linear combination of objective functions
J_mp = _obj.mse_weighted_loss(y_mp, y_mp_est, len_section=10000, weight_type="std")
J_fr_poss = _obj.poisson_weighted_loss(y_fr, y_fr_est, len_section=10000, weight_type="mean")
J_fr_mse = _obj.mse_weighted_loss(y_fr, y_fr_est, len_section=10000, weight_type="mean")
J_fr = J_fr_poss + J_fr_mse
J = J_mp + J_fr
return J
def LNKS_MP_fobj_helper(f, theta, stim, y_data, options):
'''
LNKS model objective helper function for using both membrane potential and firing rate
returns objective function value J
Inputs
-------
f: LNKS model
theta: model parameters
stim: input data
y_data: output data tuple (mp, fr)
pathway (int): LNK pathway (1 or 2)
Outputs
-------
J: objective value
'''
# data
y_mp = y_data[0]
y_fr = y_data[1]
# model output
y_mp_est, y_fr_est = f(theta, stim, options['pathway'])
# linear combination of objective functions
J_mp = _obj.mse_weighted_loss(y_mp, y_mp_est, len_section=10000, weight_type="std")
J_fr_poss = _obj.poisson_weighted_loss(y_fr, y_fr_est, len_section=10000, weight_type="mean")
J_fr_mse = _obj.mse_weighted_loss(y_fr, y_fr_est, len_section=10000, weight_type="mean")
J_fr = J_fr_poss + J_fr_mse
J = J_mp + J_fr
return J
def LNKS_fobj_helper(f, theta, stim, y, options):
'''
LNKS model objective function helper function
Weighted sum of log-likelihood and mean-square error
'''
y_est = f(theta, stim, options['pathway'])
# linear combination of objective functions
J_poss = _obj.poisson_weighted_loss(y, y_est, len_section=10000, weight_type="mean")
J_mse = _obj.mse_weighted_loss(y, y_est, len_section=10000, weight_type="mean")
J = J_poss + J_mse
return J
