plt.subplot(212)
plt.plot(pup_m,'-o', color='k', alpha=0.5, lw=2)
plt.plot(diff, '-o', color='g')
plt.legend(['pup', 'rN-r0'])
plt.xlabel('pip trials')
plt.title('corr coef btwn rN-r0 and pupil: %s'
%(np.corrcoef(diff, pup_m)[0][1]))
print('Comparing different models...')
def onpick3(event):
ind = event.ind
print('onpick3 scatter:', cellids[ind][0])
rN_perf = eval_fit(resp, fullModel)
r0_perf = eval_fit(resp, pred)
x = np.linspace(-1,1,3)
ncols = 8
nrows = 10
cellcount = resp.shape[-1]
repcount = resp.shape[1]
bincount = resp.shape[0]
spontonly=0
fs=100
stim=0
color = np.arange(cellcount)
fig = plt.figure()
for rep in range(0, repcount):
ax = fig.add_subplot(nrows,ncols,rep+1)
a_p_unwrapped = list(np.tile(a_p, (30, 1)).T.reshape(repcount * stimcount))
pupil_unwrapped = np.nanmean(pupil, 0).reshape(repcount * stimcount)
rN = NRF_fit(r=resp, r0_strf=pred, model='NRF_STRF', spontonly=False)
fig = plt_perf_by_trial(resp,
rN,
pred,
combine_stim=False,
a_p=a_p_unwrapped,
pupil=pupil,
pop_state={
'method': 'SVD',
'dims': 5
})
# Evaluate r0 and rN
cc_rN = eval_fit(resp, rN)
cc_r0 = eval_fit(resp, pred)
cc_rN_flat = cc_rN['bytrial'].reshape(repcount * stimcount, cellcount)
cc_r0_flat = cc_r0['bytrial'].reshape(repcount * stimcount, cellcount)
import statsmodels.api as sm
import pandas as pd
reg_in = pd.DataFrame(np.vstack((a_p_unwrapped, pupil_unwrapped)).T,
columns=['behavior', 'pupil'])
reg_out = pd.DataFrame(cc_rN_flat - cc_r0_flat, columns=cellids)
for cid in reg_out:
if (np.any(np.isnan(reg_out[cid]))):
for i, val in enumerate(reg_out[cid]):
if np.isnan(val):
inds = []
for ind in np.argwhere(np.isnan(pupil[0, :, :])):
inds.append(ind[0])
inds = np.array(inds)
drop_inds = np.unique(inds)
keep_inds = [x for x in np.arange(0, len(resp[0, :, 0, 0])) if x not in inds]
a_p = np.array(a_p)[keep_inds]
resp = resp[:, keep_inds, :, :]
pred = pred[:, keep_inds, :, :]
pupil = pupil[:, keep_inds, :]
pup = pupil
from NRF_tools import NRF_fit, eval_fit
rN = NRF_fit(resp, r0_strf=pred, model="NRF_STRF", spontonly=0, shuffle=True)
rN_perf = eval_fit(resp, rN)
r0_perf = eval_fit(resp, pred)
#cc_rN_all = np.empty((resp.shape[1], resp.shape[2], resp.shape[-1]))
#cc_r0_all = np.empty((resp.shape[1], resp.shape[2], resp.shape[-1]))
cc_rN_all = rN_perf['bytrial']
cc_r0_all = r0_perf['bytrial']
for i in range(0, resp.shape[1]):
for stim in range(0, resp.shape[2]):
for cell in range(0, resp.shape[-1]):
cc_rN_all[i, stim, cell] = np.corrcoef(rN[:, i, stim, cell],
resp[:, i, stim,
cell])[0][1]
cc_r0_all[i, stim, cell] = np.corrcoef(pred[:, i, stim, cell],
resp[:, i, stim,
cell])[0][1]
inds = []
for ind in np.argwhere(np.isnan(pupil[0, :, :])):
inds.append(ind[0])
inds = np.array(inds)
drop_inds = np.unique(inds)
keep_inds = [x for x in np.arange(0, len(resp[0, :, 0, 0])) if x not in inds]
a_p = np.array(a_p)[keep_inds]
resp = resp[:, keep_inds, :, :]
pred = pred[:, keep_inds, :, :]
pupil = pupil[:, keep_inds, :]
pup = pupil
from NRF_tools import NRF_fit, eval_fit
rN = NRF_fit(resp, r0_strf=pred, model="NRF_STRF", spontonly=0, shuffle=True)
rN_perf = eval_fit(resp, rN)
r0_perf = eval_fit(resp, pred)
cc_rN_all = np.empty((resp.shape[1], resp.shape[2], resp.shape[-1]))
cc_r0_all = np.empty((resp.shape[1], resp.shape[2], resp.shape[-1]))
for i in range(0, resp.shape[1]):
for stim in range(0, resp.shape[2]):
for cell in range(0, resp.shape[-1]):
cc_rN_all[i, stim, cell] = np.corrcoef(rN[:, i, stim, cell],
resp[:, i, stim,
cell])[0][1]
cc_r0_all[i, stim, cell] = np.corrcoef(pred[:, i, stim, cell],
resp[:, i, stim,
cell])[0][1]
cc_rN = np.mean(cc_rN_all[:, :, :], 1)