###################### END CUSTOMIZATION #######################################
fitness = nrd_fitness.specie_concentration_fitness(species_list=mol)
############ Test fitness function
#sim = aju.xml.NeurordSimulation('/tmp', model=model, params=params)
#cp /tmp/???/model.h5 modelname.split('.')[0]+'.h5'
#sim2=aju.xml.NeurordResult('Model_syngap_ras.h5')
#print(fitness(sim2, exp))
################
fit = aju.optimize.Fit(tmpdir, exp, model_set, None, fitness, params,
_make_simulation=aju.xml.NeurordSimulation.make,
_result_constructor=aju.xml.NeurordResult)
fit.load()
fit.do_fit(iterations, popsize=popsize,sigma=1.0,seed=seed)
mean_dict,std_dict,CV=converge.iterate_fit(fit,test_size,popsize)
########################################### Done with fitting
#to look at centroid [0] or stdev [6] of cloud of good results:
for i,p in enumerate(fit.params.unscale(fit.optimizer.result()[0])):
print(fit.param_names()[i],'=',p, '+/-', fit.params.unscale(fit.optimizer.result()[6])[i])
#to look at fit history
aju.drawing.plot_history(fit,fit.measurement)
startgood=0
threshold=20
save_params.save_params(fit, startgood, threshold)
# set-up and do the optimization
fit = fc.fit_commands(dirname,
exp_to_fit,
modeltype,
ntype,
fitness,
params1,
generations,
popsiz,
seed,
test_size,
map_func=None)
if test_size > 0:
mean_dict, std_dict, CV = converge.iterate_fit(
fit,
test_size,
popsiz,
std_crit=0.01,
slope_crit=1e-3,
max_evals=100000)
###########look at results
#from ajustador import drawing
#drawing.plot_history(fit1, fit1.measurement)
startgood = 0 #set to 0 to print all
threshold = 10 #set to large number to print all
save_params.save_params(fit1, startgood, threshold)
#save_params.persist(fit1,'.')
modeltype, ntype,
fitness, params,
_make_simulation=aju.optimize.MooseSimulation.make,
_result_constructor=aju.optimize.MooseSimulationResult)
fit.load()
fit.do_fit(generations, popsize=popsiz, seed=seed)
startgood=1000 #set to 0 to print all
threshold=0.8 #set to large number to print all
s_crt = 2E-3
max_eval = 5000
while(True):
mean_dict,std_dict,CV=converge.iterate_fit(fit,test_size,popsiz, slope_crit=s_crt, max_evals=max_eval)
save_params.save_params(fit, startgood, threshold)
char = input("plot_history opt (Y/N):")
if char.upper() == 'Y':
drawing.plot_history(fit, fit.measurement)
char = input("Continue opt (Y/N):")
if char.upper() == 'N':
break
else:
s_crt = np.float32(input("slope_criteria old_cirterial is {}?".format(s_crt)))
max_eval = np.long(input("Maximum evaluations must be > {}?".format(len(fit))))
continue
#Save parameters of good results from end of optimization, and all fitness values
#startgood=1000 #set to 0 to print all
#threshold=0.8 #set to large number to print all
#save_params.save_params(fit, startgood, threshold)
dirname = 'cond_' + dataname + str(seed) + str(popsiz)
if not dirname in os.listdir(rootdir):
os.mkdir(rootdir + dirname)
os.chdir(rootdir + dirname)
######## set up parameters and fitness to be used for all opts ############
params1, fitness = pfc.params_fitness(morph_file, ntype, modeltype)
# set-up and do the optimization
fit1 = fit_commands.fit_commands(dirname, exp_to_fit, modeltype, ntype,
fitness, params1, generations, popsiz, seed,
test_size)
if test_size > 0:
mean_dict, std_dict, CV = converge.iterate_fit(fit1,
test_size,
popsiz,
std_crit=0.02,
max_evals=12000)
###########look at results
drawing.plot_history(fit1, fit1.measurement)
#Save parameters of good results toward the end, and all fitness values
startgood = 0 #set to 0 to print/save all
threshold = 10 #set to high value to print/save all
save_params.save_params(fit1, startgood, threshold)
#save_params.persist(fit1,'.')
'''
Repeat arky120, arky140, proto079,proto122 optimizations with channel kinetics allowed to vary
but with limitations to axonal conductance parameters described in Lindroos
spike_range_y_histogram=1)
########### Neuron and fit specific commands ############
fit1 = aju.optimize.Fit(tmpdir,
exp_to_fit,
modeltype,
ntype,
fitness,
params1,
_make_simulation=aju.optimize.MooseSimulation.make,
_result_constructor=aju.optimize.MooseSimulationResult)
fit1.load()
fit1.do_fit(generations, popsize=popsiz, seed=seed)
mean_dict1, std_dict1, CV1 = converge.iterate_fit(fit1, test_size, popsiz)
#look at results
drawing.plot_history(fit1, fit1.measurement)
#Save parameters of good results toward the end, and all fitness values
startgood = 1500 #set to 0 to print all
threshold = 0.4 #set to large number to print all
save_params.save_params(fit1, startgood, threshold)
#save_params.persist(fit1,'.')
################## Next neuron #############
dataname = 'proto122'
exp_to_fit = gpe.data[dataname + '-2s'][[0, 1, 4]]
spike_ahp=1,
ahp_curve=4,
charging_curve=1,
spike_range_y_histogram=1)
########### Neuron and fit specific commands ############
fit1 = aju.optimize.Fit(tmpdir,
exp_to_fit,
modeltype,
ntype,
fitness,
params1,
_make_simulation=aju.optimize.MooseSimulation.make,
_result_constructor=aju.optimize.MooseSimulationResult)
fit1.load()
fit1.do_fit(generations, popsize=popsiz, seed=seed)
mean_dict, std_dict, CV = converge.iterate_fit(fit1, test_size, popsiz)
#look at results
drawing.plot_history(fit1, fit1.measurement)
#Save parameters of good results from end of optimization, and all fitness values
startgood = 1000 #set to 0 to print all
threshold = 0.8 #set to large number to print all
save_params.save_params(fit1, startgood, threshold)
#to save the fit object
#save_params.persist(fit1,'.')
spike_range_y_histogram=1)
########### Neuron and fit specific commands ############
fit1 = aju.optimize.Fit(tmpdir,
exp_to_fit,
modeltype,
ntype,
fitness,
params1,
_make_simulation=aju.optimize.MooseSimulation.make,
_result_constructor=aju.optimize.MooseSimulationResult)
fit1.load()
fit1.do_fit(generations, popsize=popsiz, seed=seed)
mean_dict1, std_dict1, CV1 = converge.iterate_fit(fit1, test_size, popsiz)
#look at results
drawing.plot_history(fit1, fit1.measurement)
#Save parameters of good results toward the end, and all fitness values
startgood = 1500 #set to 0 to print all
threshold = 0.40 #median
save_params.save_params(fit1, startgood, threshold)
#save_params.persist(fit1,'.')
################## Next neuron #############
dataname = 'proto154'
exp_to_fit = gpe.data[dataname + '-2s'][[0, 2, 4]]
dirname = dataname + 'F_' + str(seed)
#sim = aju.xml.NeurordSimulation('/tmp', model=model, params=params)
#sim2=aju.xml.NeurordResult('Model_syngap_ras.h5')
#print(fitness(sim2, exp))
################
fit = aju.optimize.Fit(tmpdir,
exp,
model_set,
None,
fitness,
params,
_make_simulation=aju.xml.NeurordSimulation.make,
_result_constructor=aju.xml.NeurordResult)
fit.load()
fit.do_fit(iterations, popsize=popsize, sigma=0.3)
mean_dict, std_dict, CV = converge.iterate_fit(fit, test_size, popsize)
########################################### Done with fitting
#to look at fit history
aju.drawing.plot_history(fit, fit.measurement)
#print centroid [0] and stdev [6] of cloud of good results:
for i, p in enumerate(fit.params.unscale(fit.optimizer.result()[0])):
print(fit.param_names()[i], '=', p, '+/-',
fit.params.unscale(fit.optimizer.result()[6])[i])
save_params.save_params(fit, 0, 1)
########################################## Next model
model_set = 'Model-CKnew-ss' #uses 6 parameter optimization
