def OLD_regress(density):
dcols = sorted([c for c in density.columns if isinstance(c,float)])
uchoice = density.choice.values
uvals = density[dcols].values
isnan = np.isnan(uchoice)
if np.sum(isnan)>0:
print('Excluding {:0.0f}% nans.'.format(np.mean(isnan)*100))
uchoice = uchoice[~isnan]
uvals = uvals[~isnan]
try:
reg = Logit(uchoice,uvals).fit(disp=False)
reg_params = reg.params
reg_err = np.abs(reg.conf_int(alpha=0.05).T - reg.params)
except (np.linalg.LinAlgError, sm.tools.sm_exceptions.PerfectSeparationError):
reg_params = np.nan * np.zeros(uvals.shape[1])
reg_err = np.nan * np.zeros([2, len(reg_params)])
res = pd.DataFrame(index=dcols)
res.loc[:,'weight'] = reg_params
if not np.any(np.isnan(reg_err)):
assert np.allclose(reg_err[0],reg_err[1]) # symmetrical errorbars
res.loc[:,'yerr'] = reg_err[0] # half of confidence interval
else:
res.loc[:,'yerr'] = np.nan
return res
def regress(density, nbins=None, only_dur=None, error='95ci', fit_intercept=False, r_and_l=False):
if only_dur is not None:
density = density[density.dur==only_dur]
else:
only_dur = density.dur.values[0]
assert np.all(density.dur.values == density.dur.values[0]), 'Cannot regress on different trial lengths at once.'
dcols_L = np.array(sorted([c for c in density.columns if c.startswith('L_')]))
dcols_R = np.array(sorted([c for c in density.columns if c.startswith('R_')]))
dcols_L_num = np.array([float(i[2:]) for i in dcols_L])
dcols_R_num = np.array([float(i[2:]) for i in dcols_L])
# restrict to bins that exist for duration of interest
is_indur_L = dcols_L_num <= only_dur
is_indur_R = dcols_R_num <= only_dur
dcols_L = dcols_L[is_indur_L]
dcols_R = dcols_R[is_indur_R]
dcols_L_num = dcols_L_num[is_indur_L]
dcols_R_num = dcols_R_num[is_indur_R]
assert np.all(dcols_L_num == dcols_R_num)
dcols_num = dcols_L_num
uchoice = density.choice.values
udens_L = density[dcols_L].values
udens_R = density[dcols_R].values
udens = udens_R-udens_L
if nbins is not None:
udens = downsample_bins(udens, nbins)
udens_L = downsample_bins(udens_L, nbins)
udens_R = downsample_bins(udens_R, nbins)
dcols_num = downsample_bins(dcols_num, nbins, method=np.mean)
isnan = np.isnan(uchoice)
if np.sum(isnan)>0:
print('Excluding {:0.0f}% nans.'.format(np.mean(isnan)*100))
uchoice = uchoice[~isnan]
udens = udens[~isnan]
udens_L = udens_L[~isnan]
udens_R = udens_R[~isnan]
if r_and_l: # use r and l as separate regressors
udens = np.concatenate([udens_L, udens_R], axis=1)
if fit_intercept:
udens = sm.add_constant(udens)
try:
reg = Logit(uchoice, udens, missing='drop').fit(disp=False)
reg_params = reg.params
if error == '95ci':
reg_err = np.abs(reg.conf_int(alpha=0.05).T - reg.params) # 95% CI
elif error == '99ci':
reg_err = np.abs(reg.conf_int(alpha=0.01).T - reg.params) # 99% CI
elif error == 'se':
reg_err = reg.bse # standard error
elif error == 'bootstrap':
boots = []
for i in range(1000):
samp = np.random.choice(np.arange(len(udens)), replace=True, size=len(udens))
udb = udens[samp]
uch = uchoice[samp]
boots.append(Logit(uch, udb, missing='drop').fit(disp=False).params)
reg_err = np.std(boots, axis=0) #/ np.sqrt(len(boots))
except (np.linalg.LinAlgError, sm.tools.sm_exceptions.PerfectSeparationError):
reg_params = np.nan * np.zeros(udens.shape[1])
reg_err = np.nan * np.zeros([2, len(reg_params)])
index = dcols_num
if r_and_l:
index = np.tile(index, 2)
if fit_intercept:
index = np.append(-1, index)
res = pd.DataFrame(index=index)
res.loc[:,'weight'] = reg_params
if not np.any(np.isnan(reg_err)):
if reg_err.ndim == 2:
assert np.allclose(reg_err[0],reg_err[1]) # symmetrical errorbars
res.loc[:,'yerr'] = reg_err[0] # half of confidence interval
elif reg_err.ndim == 1:
res.loc[:,'yerr'] = reg_err
else:
res.loc[:,'yerr'] = np.nan
if r_and_l:
resl = res.iloc[:len(res)//2]
resl.columns = [c+'_L' for c in resl.columns]
resr = res.iloc[len(res)//2:]
resr.columns = [c+'_R' for c in resr.columns]
res = resl.join(resr)
return res
