def generate_prediction(est, val, modelspecs):
list_val = False
if type(val) is list:
# ie, if jackknifing
list_val = True
else:
# Evaluate estimation and validation data
# Since ms.evaluate only does a shallow copy of rec, successive
# evaluations of one rec on many modelspecs just results in a list of
# different pointers to the same recording. So need to force copies of
# est/val before evaluating.
if len(modelspecs) == 1:
# no copies needed for 1 modelspec
est = [est]
val = [val]
else:
est = [est.copy() for i, _ in enumerate(modelspecs)]
val = [val.copy() for i, _ in enumerate(modelspecs)]
new_est = [ms.evaluate(d, m) for m, d in zip(modelspecs, est)]
new_val = [ms.evaluate(d, m) for m, d in zip(modelspecs, val)]
if list_val:
new_val = [recording.jackknife_inverse_merge(new_val)]
return new_est, new_val
def generate_prediction(est, val, modelspecs):
if type(val) is list:
# ie, if jackknifing
new_est = [ms.evaluate(d, m) for m, d in zip(modelspecs, est)]
new_val = [ms.evaluate(d, m) for m, d in zip(modelspecs, val)]
new_val = [recording.jackknife_inverse_merge(new_val)]
else:
# Evaluate estimation and validation data
new_est = [ms.evaluate(est, m) for m in modelspecs]
new_val = [ms.evaluate(val, m) for m in modelspecs]
return new_est, new_val
def generate_prediction(est, val, modelspecs):
list_val = False
if type(val) is list:
# ie, if jackknifing
list_val = True
else:
# Evaluate estimation and validation data
# Since ms.evaluate only does a shallow copy of rec, successive
# evaluations of one rec on many modelspecs just results in a list of
# different pointers to the same recording. So need to force copies of
# est/val before evaluating.
if len(modelspecs) == 1:
# no copies needed for 1 modelspec
est = [est]
val = [val]
else:
est = [est.copy() for i, _ in enumerate(modelspecs)]
val = [val.copy() for i, _ in enumerate(modelspecs)]
new_est = []
new_val = []
for m, e, v in zip(modelspecs, est, val):
# nan-out periods outside of mask
e = ms.evaluate(e, m)
v = ms.evaluate(v, m)
if 'mask' in v.signals.keys():
m = v['mask'].as_continuous()
x = v['pred'].as_continuous().copy()
x[..., m[0,:] == 0] = np.nan
v['pred'] = v['pred']._modified_copy(x)
new_est.append(e)
new_val.append(v)
#new_est = [ms.evaluate(d, m) for m, d in zip(modelspecs, est)]
#new_val = [ms.evaluate(d, m) for m, d in zip(modelspecs, val)]
if list_val:
new_val = [recording.jackknife_inverse_merge(new_val)]
return new_est, new_val
def stim_estimator_PCA(rec, npcs):
x = len(rec['resp'].chans)
##########
pca_out = PCA(rec['resp'], center=False)
rec['resp'] = rec['resp'].rasterize()._modified_copy(
np.matmul(pca_out['pcs'][:, 0:npcs], pca_out['loading'][0:npcs, :]).T)
for i in range(10):
rec['stim'] = rec['stim'].rasterize()
i += 1
nfolds = 10
ests, vals, m = preproc.mask_est_val_for_jackknife(rec, modelspecs=None,
njacks=nfolds)
for i in range(10):
vals[i]['stim'] = vals[i]['stim'].rasterize()
i += 1
new_val = recording.jackknife_inverse_merge(vals)
S_est_list = []
for i in range(10):
est = ests[i].apply_mask()
stim = est['stim']
resp_est = est['resp'].rasterize()
X_est = stim.rasterize().as_continuous()
Y_est = resp_est.as_continuous()
mean_array = dc.mean_array_resp(x, Y_est)
Y_est_subtracted = dc.Y_est_subtractedavg(mean_array, x, Y_est)
S = X_est # stim est set
R = Y_est_subtracted # response est set
R1 = np.concatenate((R[:x, 2:], np.zeros((x, 2))), axis=1)
R2 = np.concatenate((R, R1), axis=0)
R3 = np.concatenate((R[:x, 4:], np.zeros((x, 4))), axis=1)
R4 = np.concatenate((R2, R3), axis=0)
R5 = np.concatenate((R[:x, 6:], np.zeros((x, 6))), axis=1)
R6 = np.concatenate((R4, R5), axis=0)
R7 = np.concatenate((R[:x, 8:], np.zeros((x, 8))), axis=1)
R8 = np.concatenate((R6, R7), axis=0)
R9 = np.concatenate((R[:x, 10:], np.zeros((x, 10))), axis=1)
R10 = np.concatenate((R8, R9), axis=0)
R11 = np.concatenate((R[:x, 12:], np.zeros((x, 12))), axis=1)
R12 = np.concatenate((R10, R11), axis=0)
R13 = np.concatenate((R[:x, 14:], np.zeros((x, 14))), axis=1)
R14 = np.concatenate((R12, R13), axis=0)
R15 = np.concatenate((R[:x, 16:], np.zeros((x, 16))), axis=1)
R16 = np.concatenate((R14, R15), axis=0)
R17 = np.concatenate((R[:x, 18:], np.zeros((x, 18))), axis=1)
R18 = np.concatenate((R16, R17), axis=0)
R19 = np.concatenate((R[:x, 20:], np.zeros((x, 20))), axis=1)
R20 = np.concatenate((R18, R19), axis=0)
R_trans = np.transpose(R20)
Crr = np.matmul(R20, R_trans)
Crs = np.matmul(S, R_trans)
Crr_inv = np.linalg.inv(Crr)
G = np.matmul(Crs, Crr_inv)
resp_val = new_val['resp']
Y_val = resp_val.as_continuous()
Y_val = np.nan_to_num(Y_val)
shape = Y_val.shape
y = shape[1]
Y_val_segment = Y_val[:, int((y / 10) * i):int((y / 10) * (i + 1))]
shape = Y_val_segment.shape
y = shape[1]
Y_val_subtractedavg = np.zeros(shape=(x, y))
for u in range(x):
Y_est_cell_mean = mean_array[u]
difference = Y_val_segment[u, :] - Y_est_cell_mean
Y_val_subtractedavg[u] = difference
u += 1
R_val = Y_val_subtractedavg
R1_val = np.concatenate((R_val[:x, 2:], np.zeros((x, 2))), axis=1)
R2_val = np.concatenate((R_val, R1_val), axis=0)
R3_val = np.concatenate((R_val[:x, 4:], np.zeros((x, 4))), axis=1)
R4_val = np.concatenate((R2_val, R3_val), axis=0)
R5_val = np.concatenate((R_val[:x, 6:], np.zeros((x, 6))), axis=1)
R6_val = np.concatenate((R4_val, R5_val), axis=0)
R7_val = np.concatenate((R_val[:x, 8:], np.zeros((x, 8))), axis=1)
R8_val = np.concatenate((R6_val, R7_val), axis=0)
R9_val = np.concatenate((R_val[:x, 10:], np.zeros((x, 10))), axis=1)
R10_val = np.concatenate((R8_val, R9_val), axis=0)
R11_val = np.concatenate((R_val[:x, 12:], np.zeros((x, 12))), axis=1)
R12_val = np.concatenate((R10_val, R11_val), axis=0)
R13_val = np.concatenate((R_val[:x, 14:], np.zeros((x, 14))), axis=1)
R14_val = np.concatenate((R12_val, R13_val), axis=0)
R15_val = np.concatenate((R_val[:x, 16:], np.zeros((x, 16))), axis=1)
R16_val = np.concatenate((R14_val, R15_val), axis=0)
R17_val = np.concatenate((R_val[:x, 18:], np.zeros((x, 18))), axis=1)
R18_val = np.concatenate((R16_val, R17_val), axis=0)
R19_val = np.concatenate((R_val[:x, 20:], np.zeros((x, 20))), axis=1)
R20_val = np.concatenate((R18_val, R19_val), axis=0)
S_est = np.matmul(G, R20_val)
S_est[S_est < 0] = 0
S_est_list.append(S_est)
i += 1
S_est = np.concatenate((S_est_list[0], S_est_list[1], S_est_list[2], S_est_list[3], S_est_list[4], S_est_list[5],
S_est_list[6], S_est_list[7], S_est_list[8], S_est_list[9]), axis=1)
return S_est