def fn(x):
summed = np.expand_dims(np.sum(x, axis=0), axis=0)
if not np.all(offsets == 0):
summed = _offset_coefficients(summed,
offsets=offsets,
fs=fs,
pad_bins=False)
return summed
def _get_fir_coefficients(modelspec, idx=0, fs=None):
i = 0
for m in modelspec:
if 'fir' in m['fn']:
if 'fn_coefficients' in m.keys():
fn = ms._lookup_fn_at(m['fn_coefficients'])
kwargs = {**m['fn_kwargs'], **m['phi']} # Merges dicts
return fn(**kwargs)
#elif 'pole_zero' in m['fn']:
# c = pz_coefficients(poles=m['phi']['poles'],
# zeros=m['phi']['zeros'],
# delays=m['phi']['delays'],
# gains=m['phi']['gains'],
# n_coefs=m['fn_kwargs']['n_coefs'], fs=100)
# return c
elif 'dexp' in m['fn']:
c = fir_dexp_coefficients(phi=m['phi']['phi'],
n_coefs=m['fn_kwargs']['n_coefs'])
return c
elif 'exp' in m['fn']:
tau = m['phi']['tau']
if 'a' in m['phi']:
a = m['phi']['a']
else:
a = m['fn_kwargs']['a']
if 'b' in m['phi']:
b = m['phi']['b']
else:
b = m['fn_kwargs']['b']
c = fir_exp_coefficients(tau,
a=a,
b=b,
n_coefs=m['fn_kwargs']['n_coefs'])
return c
elif i == idx:
coefficients = m['phi']['coefficients']
if 'offsets' in m['phi']:
if fs is None:
log.warning("couldn't compute offset coefficients for "
"STRF heatmap, no fs provided.")
else:
coefficients = _offset_coefficients(
coefficients, m['phi']['offsets'], fs=fs)
return coefficients
else:
i += 1
return None
def contrast(rec,
tau,
a,
b,
s,
mean,
sd,
i='stim',
o='ctpred',
c='contrast',
offsets=0.0,
n_channels=18,
n_coefs=15,
compute_contrast=False):
if compute_contrast:
wc_coeffs = gaussian_coefficients(mean, sd, n_channels)
fir_coeffs = fir_exp_coefficients(tau, a, b, s, n_coefs=15)
if not np.all(offsets == 0):
fs = rec[i].fs
fir_coeffs = _offset_coefficients(fir_coeffs,
offsets,
fs,
pad_bins=True)
wc_coeffs = np.abs(wc_coeffs)
fir_coeffs = np.abs(fir_coeffs)
def fn(x):
weighted = wc_coeffs.T * x
weighted[np.isnan(weighted)] = 0
width = wc_coeffs.shape[0]
history = fir_coeffs.shape[-1]
zero_pad = np.zeros([width, history - 1])
filt = np.concatenate((zero_pad, fir_coeffs), axis=1)
filt /= np.sum(fir_coeffs)
mn = convolve2d(weighted, filt, mode='same', fillvalue=0)
var = convolve2d(weighted**2, filt, mode='same',
fillvalue=0) - mn**2
ct = np.sqrt(var) / (mn * .99 + np.nanmax(mn) * 0.01)
ctpred = np.sum(ct, axis=0)
ctpred = np.expand_dims(ctpred, axis=0)
return ctpred
else:
# pass through zeros until ready to fit GC portion of the model
fn = lambda x: np.zeros((1, x.shape[-1]))
return [rec[i].transform(fn, o)]
def contrast_kernel_heatmap2(rec,
modelspec,
ax=None,
title=None,
idx=0,
channels=0,
xlabel='Lag (s)',
ylabel='Channel In',
**options):
ct_idx = nu.find_module('contrast', modelspec)
phi = copy.deepcopy(modelspec[ct_idx]['phi'])
fn_kwargs = copy.deepcopy(modelspec[ct_idx]['fn_kwargs'])
fs = rec['stim'].fs
wc_kwargs = {k: phi[k] for k in ['mean', 'sd']}
wc_kwargs['n_chan_in'] = fn_kwargs['n_channels']
fir_kwargs = {k: phi[k] for k in ['tau', 'a', 'b', 's']}
fir_kwargs['n_coefs'] = fn_kwargs['n_coefs']
wc_coefs = gaussian_coefficients(**wc_kwargs)
fir_coefs = fir_exp_coefficients(**fir_kwargs)
if 'offsets' in phi:
offsets = phi['offsets']
elif 'offsets' in fn_kwargs:
offsets = fn_kwargs['offsets']
else:
offsets = None
if offsets is not None:
fir_coefs = _offset_coefficients(fir_coefs, offsets, fs, pad_bins=True)
wc_coefs = np.abs(wc_coefs).T
fir_coefs = np.abs(fir_coefs)
strf = wc_coefs @ fir_coefs
# TODO: This isn't really doing the same operation as an STRF anymore
# so it may be better not to plot it this way in the future.
_strf_heatmap(strf,
wc_coefs,
fir_coefs,
xlabel=xlabel,
ylabel=ylabel,
ax=ax,
title=title)
return ax
def _get_ctk_coefficients(wc_coefficients=None,
fir_coefficients=None,
mean=None,
sd=None,
coefficients=None,
f1s=None,
taus=None,
delays=None,
gains=None,
use_phi=False,
n_coefs=18,
offsets=None,
offset=None,
fs=None,
**kwargs):
if use_phi:
wc_coeffs = gaussian_coefficients(mean, sd, n_coefs)
if coefficients is None:
fir_coeffs = do_coefficients(f1s=f1s,
taus=taus,
delays=delays,
gains=gains,
n_coefs=n_coefs)
else:
fir_coeffs = coefficients
else:
if (wc_coefficients is None) or (fir_coefficients is None):
raise ValueError("contrast_kernel module was called without "
"wc or fir coefficients set.")
wc_coeffs = wc_coefficients
fir_coeffs = fir_coefficients
if (offsets is not None) and (fs is not None):
fir_coeffs = _offset_coefficients(fir_coeffs, offsets, fs)
elif (offsets is None) and (offset is not None):
# old model
offset = int(offset)
fir_coeffs = np.concatenate((np.zeros(
(fir_coeffs.shape[0], offset)), fir_coeffs[:, :-1 * offset]),
axis=1)
coeffs = np.abs(wc_coeffs.T @ fir_coeffs)
return coeffs, wc_coeffs.T, fir_coeffs
def offset_coefficients(self, coefficients, offsets, fs, pad_bins):
return _offset_coefficients(coefficients, offsets, fs, pad_bins=pad_bins)