def run_models(mname, project, regress=False):
#bayes fit all subject
allsx_df = fit_sx(mname, project=project, regress=regress)
#parse model output
subdf = parse_allsx(allsx_df)
pdict = subdf_to_pdict(subdf)
#simulate and compare with observed data
data = defmod.find_data(mname, project=project)
simdf = vis.predict(pdict,
data,
ntrials=160,
nsims=100,
save=True,
RTname="SimRT_EvT.jpeg",
ACCname="SimACC_EvT.jpeg")
simdf.to_csv("simdf_sxbayes.csv")
#save pdict; can be reloaded and transformed back into
#the original pdict format by the following commands
#1. pdict=pd.read_csv("sxbayes_pdict.csv")
#2. pdict=pdict.to_dict()
params = pd.DataFrame(pdict)
params.to_csv("sxbayes_pdict.csv", index=False)
def optimize_sx(mname, project='imaging'):
m = defmod.define_model(mname, project=project)
data = m.data
if 'z' in m.depends_on.keys():
bias = True
else:
bias = False
grp_dict = {}
subj_params = []
aic_list = []
bic_list = []
dic_list = []
ic_dict = {}
for subj_idx, subj_data in data.groupby('subj_idx'):
m_subj = hddm.HDDM(subj_data,
depends_on=m.depends_on,
bias=bias,
include=m.include)
sx_params = m_subj.optimize('ML')
pdict = opt.get_pdict(sx_params)
subj_params.append(sx_params)
aic_list.append(aic(m_subj))
bic_list.append(bic(m_subj))
#dic_list.append(m_subj.dic)
grp_dict[subj_idx] = pdict
ic_dict = {'aic': aic_list, 'bic': bic_list}
ic_df = pd.DataFrame(ic_dict)
ic_df.to_csv(mname + "_IC_Rank.csv")
#write grp_dict to .txt file for reloading later
f = open('mle_params.txt', 'w')
f.write('grp_dict=' + repr(grp_dict) + '\n')
f.close()
params = pd.DataFrame(subj_params)
simdf = vis.predict(grp_dict,
data,
ntrials=100,
nsims=100,
save=True,
RTname="dbmz_RT.jpeg",
ACCname="dbmz_ACC.jpeg")
#simdf=vis.predict(grp_dict, df, ntrials=160, nsims=100, save=True, RTname="SimRT_EvT.jpeg", ACCname="SimACC_EvT.jpeg")
simdf.to_csv("simdf_opt.csv")
params.to_csv("subj_params_opt.csv", index=False)
sdt.plot_rho_sdt(data, simdf)
empvsim = sdt.rho_sdt(data, simdf)
return grp_dict, ic_df
def run_models(mname, project, regress=False):
#bayes fit all subject
allsx_df=fit_sx(mname, project=project, regress=regress)
#parse model output
subdf=parse_allsx(allsx_df)
pdict=subdf_to_pdict(subdf)
#simulate and compare with observed data
data=defmod.find_data(mname, project=project)
simdf=vis.predict(pdict, data, ntrials=160, nsims=100, save=True, RTname="SimRT_EvT.jpeg", ACCname="SimACC_EvT.jpeg")
simdf.to_csv("simdf_sxbayes.csv")
#save pdict; can be reloaded and transformed back into
#the original pdict format by the following commands
#1. pdict=pd.read_csv("sxbayes_pdict.csv")
#2. pdict=pdict.to_dict()
params=pd.DataFrame(pdict)
params.to_csv("sxbayes_pdict.csv", index=False)
def optimize_sx(mname, project='imaging'):
m=defmod.define_model(mname, project=project)
data=m.data
if 'z' in m.depends_on.keys():
bias=True
else:
bias=False
grp_dict={}; subj_params=[]; aic_list=[]; bic_list=[]; dic_list=[]; ic_dict={}
for subj_idx, subj_data in data.groupby('subj_idx'):
m_subj=hddm.HDDM(subj_data, depends_on=m.depends_on, bias=bias, include=m.include)
sx_params=m_subj.optimize('ML')
pdict=opt.get_pdict(sx_params)
subj_params.append(sx_params)
aic_list.append(aic(m_subj)); bic_list.append(bic(m_subj)); #dic_list.append(m_subj.dic)
grp_dict[subj_idx]=pdict
ic_dict={'aic':aic_list, 'bic':bic_list}
ic_df=pd.DataFrame(ic_dict)
ic_df.to_csv(mname+"_IC_Rank.csv")
#write grp_dict to .txt file for reloading later
f=open('mle_params.txt', 'w')
f.write('grp_dict=' + repr(grp_dict) + '\n')
f.close()
params = pd.DataFrame(subj_params)
simdf=vis.predict(grp_dict, data, ntrials=100, nsims=100, save=True, RTname="dbmz_RT.jpeg", ACCname="dbmz_ACC.jpeg")
#simdf=vis.predict(grp_dict, df, ntrials=160, nsims=100, save=True, RTname="SimRT_EvT.jpeg", ACCname="SimACC_EvT.jpeg")
simdf.to_csv("simdf_opt.csv")
params.to_csv("subj_params_opt.csv", index=False)
sdt.plot_rho_sdt(data, simdf)
empvsim=sdt.rho_sdt(data, simdf)
return grp_dict, ic_df
def postproc(m, mname='model', varlvl='grp', traces_info=None):
if not os.path.isdir("avg_pdf"):
os.makedirs("avg_pdf")
if not os.path.isdir("avg_quant"):
os.makedirs("avg_quant")
if not os.path.isdir("posteriors"):
os.makedirs("posteriors")
if not os.path.isdir("correlations"):
os.makedirs("correlations")
if not os.path.isdir("simulations"):
os.makedirs("simulations")
os.chdir("posteriors")
m.plot_posteriors(save=True)
plt.close('all')
os.chdir("../")
m.print_stats(fname=mname+'_stats.txt')
parsed = parse.parse_stats(minput=m, varlvl=varlvl, input_isdf=False)
subdf=parsed[0]
condsdf=parsed[1]
pdict=parsed[2]
os.chdir('simulations')
#pdict=mle.optimize_sx(mname)
print "simulating from mle optimized params"
simdf=vis.predict(params=pdict, data=m.data, simfx=sims.sim_exp, ntrials=240, pslow=0.0, pfast=0.0, nsims=100, save=True, RTname=mname+'_RT_simexp', ACCname=mname+'_ACC_simexp')
simdf.to_csv("simdf.csv", index=False)
simdf=pd.read_csv("simdf.csv")
vis.predict_from_simdfs(m.data, simdf)
plt.close('all')
os.chdir('../')
simdf.to_csv("simdf_opt.csv")
os.chdir('correlations')
print "estimating emp-theo sdt correlation, saving..."
sdt_corr=sdt.rho_sdt(m.data, simdf)
print "plotting emp-theo sdt correlation, saving..."
sdt.plot_evs_corr(m.data, simdf)
sdt_corr.to_csv("sdt_corr.csv")
plt.close('all')
os.chdir('../')
#mname, traces_name=defmod.find_traces_imaging(m)
avgm=m.get_average_model()
#avgm.load_db(dbname=traces_name, db='pickle')
avgm.load_db(dbname="msmIO_traces.db", db='pickle')
os.chdir("./avg_pdf")
avgm.plot_posterior_predictive(save=True, value_range=np.linspace(-6, 6, 100), figsize=(12, 10))
os.chdir("../avg_quant")
avgm.plot_posterior_quantiles(save=True, samples=250, figsize=(12, 10))
plt.close('all')