def _strf_to_contrast(modelspec, absolute_value=True):
'''
Copy prefitted WC and FIR phi values to contrast-based counterparts.
'''
modelspec = copy.deepcopy(modelspec)
wc_idx, ctwc_idx = find_module('weight_channels',
modelspec,
find_all_matches=True)
fir_idx, ctfir_idx = find_module('fir', modelspec, find_all_matches=True)
log.info("Updating contrast phi to match prefitted strf ...")
modelspec[ctwc_idx]['phi'] = copy.deepcopy(modelspec[wc_idx]['phi'])
modelspec[ctfir_idx]['phi'] = copy.deepcopy(modelspec[fir_idx]['phi'])
if absolute_value:
for k, v in modelspec[ctwc_idx]['phi'].items():
p = np.abs(v)
modelspec[ctwc_idx]['phi'][k] = p
for k, v in modelspec[ctfir_idx]['phi'].items():
p = np.abs(v)
modelspec[ctfir_idx]['phi'][k] = p
return modelspec
def fixed_contrast_strf(modelspec=None, **kwargs):
if modelspec is None:
pass
else:
# WARNING: This modifies modelspec in-place mid-evaluation.
# Really not sure this is the right way to do this.
wc_idx = find_module('weight_channels', modelspec)
if 'g' not in modelspec[wc_idx]['id']:
_, ctwc_idx = find_module('weight_channels',
modelspec,
find_all_matches=True)
fir_idx, ctfir_idx = find_module('fir',
modelspec,
find_all_matches=True)
modelspec[ctwc_idx]['fn_kwargs'].update(
copy.deepcopy(modelspec[wc_idx]['phi']))
modelspec[ctfir_idx]['fn_kwargs'].update(
copy.deepcopy(modelspec[fir_idx]['phi']))
modelspec[ctwc_idx]['phi'] = {}
modelspec[ctfir_idx]['phi'] = {}
for k, v in modelspec[ctwc_idx]['phi']:
p = np.abs(v)
modelspec[ctwc_idx]['phi'][k] = p
for k, v in modelspec[ctfir_idx]['phi']:
p = np.abs(v)
modelspec[ctfir_idx]['phi'][k] = p
return False
def _set_nonlinearity(modelspec):
# ctx = get_default_ctx()
# est = ctx['est']
# val = ctx['val']
# eresp = est['resp'].as_continuous()
# vresp = val['resp'].as_continuous()
# min_resp = min(eresp.min(), vresp.min())
# max_resp = max(eresp.max(), vresp.max())
# epred = est['pred'].as_continuous()
# vpred = val['pred'].as_continuous()
# predrange = 2/(max(epred.max() - epred.min(),
# vpred.max() - vpred.min()) + 1)
#base = np.array([min_resp])
#amplitude = np.array([max_resp*0.5])
#shift = np.array([0.5*(epred.mean() + vpred.mean())])
#kappa = np.array([np.log(predrange)])
base = np.array([0])
amplitude = np.array([2])
shift = np.array([0.275])
kappa = np.array([2.5])
dexp_idx = find_module('double_exponential', modelspec)
if dexp_idx is None:
# no dexp, assume dsig for gc instead
dexp_idx = find_module('dynamic_sigmoid', modelspec)
modelspec[dexp_idx]['phi'].update({
'base': base, 'amplitude': amplitude,
'shift': shift, 'kappa': kappa
})
return modelspec
def _set_LN_phi(modelspec):
wc_idx2 = find_module('weight_channels', modelspec)
fir_idx2 = find_module('fir', modelspec)
modelspec[wc_idx2]['phi'] = {
'mean': np.array([0.4, 0.5]),
'sd': np.array([0.15, 0.15])
}
modelspec[fir_idx2]['phi'] = {
'coefficients': np.array([
[0, -.125, -.25, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, .275, .15, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
])
}
return _set_nonlinearity(modelspec)
def add_gc_signal(rec, modelspec, name='GC'):
modelspec = copy.deepcopy(modelspec)
rec = copy.deepcopy(rec)
dsig_idx = find_module('dynamic_sigmoid', modelspec)
# if dsig_idx is None:
# log.warning("No dsig module was found, can't add GC signal.")
# return rec
phi = modelspec[dsig_idx]['phi']
phi.update(modelspec[dsig_idx]['fn_kwargs'])
pred = rec['pred'].as_continuous()
b = phi['base'] + (phi['base_mod'] - phi['base']) * pred
a = phi['amplitude'] + (phi['amplitude_mod'] - phi['amplitude']) * pred
s = phi['shift'] + (phi['shift_mod'] - phi['shift']) * pred
k = phi['kappa'] + (phi['kappa_mod'] - phi['kappa']) * pred
array = np.squeeze(np.stack([b, a, s, k], axis=0))
fs = rec['stim'].fs
signal = nems.signal.RasterizedSignal(fs,
array,
name,
rec['stim'].recording,
chans=['B', 'A', 'S', 'K'],
epochs=rec['stim'].epochs)
rec[name] = signal
return rec
def get_module(ctx, val, key='index', mspec_idx=0, find_all_matches=False):
mspec = ctx['modelspecs'][mspec_idx]
if key in ['index', 'idx', 'i']:
return mspec[val]
else:
i = find_module(val, mspec, find_all_matches=find_all_matches, key=key)
return mspec[i]
def fit_to_simulation(fit_model, simulation_spec):
'''
Parameters:
-----------
fit_model : str
Modelname to fit to the simulation.
simulation_spec : NEMS ModelSpec
Modelspec to base simulation on.
Returns:
--------
ctx : dict
Xforms context. See nems.xforms.
'''
rec = get_default_ctx()['rec']
ctk_idx = find_module('contrast_kernel', simulation_spec)
if ctk_idx is not None:
simulation_spec[ctk_idx]['fn_kwargs']['evaluate_contrast'] = True
new_resp = simulation_spec.evaluate(rec)['pred']
rec['resp'] = new_resp
# replace ozgf and ld with ldm
modelname = '-'.join(fit_model.split('-')[2:])
xfspec = xhelp.generate_xforms_spec(modelname=modelname)
ctx, _ = xforms.evaluate(xfspec, context={'rec': rec})
return ctx
def build_toy_combined_cell(base, amplitude, shift, kappa, u, tau):
modelspec = from_keywords(combined.split('_')[1])
modelspec = _set_LN_phi(modelspec)
stp_idx = find_module('stp', modelspec)
modelspec[stp_idx]['phi'] = {'u': u, 'tau': tau}
return _set_gc_phi(modelspec, base, amplitude, shift, kappa)
def dsig_phi_to_prior(modelspec):
'''
Sets priors for dynamic_sigmoid equal to the current phi for the
same module. Used for random-sample fits - all samples are initialized
and pre-fit the same way, and then randomly sampled from the new priors.
Parameters
----------
modelspec : list of dictionaries
A NEMS modelspec containing, at minimum, a dynamic_sigmoid module
Returns
-------
modelspec : A copy of the input modelspec with priors updated.
'''
modelspec = copy.deepcopy(modelspec)
dsig_idx = find_module('dynamic_sigmoid', modelspec)
dsig = modelspec[dsig_idx]
phi = dsig['phi']
b = phi['base']
a = phi['amplitude']
k = phi['kappa']
s = phi['shift']
p = dsig['prior']
p['base'][1]['beta'] = b
p['amplitude'][1]['beta'] = a
p['shift'][1]['mean'] = s # Do anything to scale sd?
p['kappa'][1]['beta'] = k
return modelspec
def _prefit_dsig_only(est,
modelspec,
analysis_function,
fitter,
metric=None,
fit_kwargs={}):
'''
Perform a rough fit that only allows dynamic_sigmoid parameters to vary.
'''
dsig_idx = find_module('dynamic_sigmoid', modelspec)
# freeze all non-static dynamic sigmoid parameters
dynamic_phi = {
'amplitude_mod': False,
'base_mod': False,
'kappa_mod': False,
'shift_mod': False
}
for p in dynamic_phi:
v = modelspec[dsig_idx]['prior'].pop(p, False)
if v:
modelspec[dsig_idx]['fn_kwargs'][p] = np.nan
dynamic_phi[p] = v
# Remove ctwc, ctfir, and ctlvl if they exist
temp = []
for i, m in enumerate(modelspec.modules):
if 'ct' in m['id']:
pass
else:
temp.append(m)
temp = ms.ModelSpec(raw=[temp])
temp = prefit_mod_subset(est,
temp,
analysis_function,
fit_set=['dynamic_sigmoid'],
fitter=fitter,
metric=metric,
fit_kwargs=fit_kwargs)
# Put ctwc, ctfir, and ctlvl back in where applicable
j = 0
for i, m in enumerate(modelspec.modules):
if 'ct' in m['id']:
pass
else:
modelspec[i] = temp[j]
j += 1
# reset dynamic sigmoid parameters if they were frozen
for p, v in dynamic_phi.items():
if v:
prior = priors._tuples_to_distributions({p: v})[p]
modelspec[dsig_idx]['fn_kwargs'].pop(p, None)
modelspec[dsig_idx]['prior'][p] = v
modelspec[dsig_idx]['phi'][p] = prior.mean()
return modelspec
def _figure_out_mod_split(modelspec):
"""
determine where to split modelspec for pop vs. slice fit
:param modelspec:
:return:
"""
bank_mod = find_module('filter_bank', modelspec, find_all_matches=True)
wc_mod = find_module('weight_channels', modelspec, find_all_matches=True)
if len(wc_mod) >= 2:
fit_set_all = list(range(wc_mod[1]))
fit_set_slice = list(range(wc_mod[1], len(modelspec)))
elif len(bank_mod) == 1:
fit_set_all = list(range(bank_mod[0]))
fit_set_slice = list(range(bank_mod[0], len(modelspec)))
else:
raise ValueError("Can't figure out how to split all and slices")
return fit_set_all, fit_set_slice
def build_toy_stp_cell(u, tau):
if not isinstance(u, np.ndarray):
u = np.array(u)
if not isinstance(tau, np.ndarray):
tau = np.array(tau)
modelspec = from_keywords(stp.split('_')[1])
modelspec = _set_LN_phi(modelspec)
stp_idx = find_module('stp', modelspec)
modelspec[stp_idx]['phi'] = {'u': u, 'tau': tau}
return modelspec
def fir_L2_norm(modelspec):
modelspec = copy.deepcopy(modelspec)
fir_idx = find_module('fir', modelspec)
prior = priors._tuples_to_distributions(modelspec[fir_idx]['prior'])
random_coeffs = np.random.rand(*prior['coefficients'].mean().shape)
normed = random_coeffs / np.linalg.norm(random_coeffs)
# Assumes fir phi hasn't been initialized yet and that coefficients
# is the only parameter to set. MAY NOT BE TRUE FOR SOME MODELS.
modelspec[fir_idx]['phi'] = {'coefficients': normed}
return modelspec
def gd_ratio(cellid, batch, modelname):
xfspec, ctx = xhelp.load_model_xform(cellid,
batch,
modelname,
eval_model=False)
mspec = ctx['modelspec']
dsig_idx = find_module('dynamic_sigmoid', mspec)
phi = mspec[dsig_idx]['phi']
return phi['kappa_mod'] / phi['kappa']
def freeze_dsig_statics(modelspec):
modelspec = copy.deepcopy(modelspec)
dsig_idx = find_module('dynamic_sigmoid', modelspec)
if dsig_idx is None:
log.warning("No dsig module was found, can't initialize.")
return modelspec
p = modelspec[dsig_idx]['phi']
frozen_bounds = {k: (v, v) for k, v in p.items()}
modelspec[dsig_idx]['bounds'].update(frozen_bounds)
return modelspec
def dsig_phi_to_prior(modelspec):
'''
Sets priors for dynamic_sigmoid equal to the current phi for the
same module. Used for random-sample fits - all samples are initialized
and pre-fit the same way, and then randomly sampled from the new priors.
Operates on modelspec IN-PLACE.
Parameters
----------
modelspec : list of dictionaries
A NEMS modelspec containing, at minimum, a dynamic_sigmoid module
Returns
-------
modelspec : A copy of the input modelspec with priors updated.
'''
dsig_idx = find_module('dynamic_sigmoid', modelspec)
phi = modelspec[dsig_idx]['phi']
b = phi['base']
a = phi['amplitude']
k = phi['kappa']
s = phi['shift']
b_m = 'base_mod' in phi
a_m = 'amplitude_mod' in phi
k_m = 'kappa_mod' in phi
s_m = 'shift_mod' in phi
amp_prior = ('Normal', {'mean': a, 'sd': np.abs(a * 2)})
base_prior = ('Exponential', {'beta': b})
kappa_prior = ('Normal', {'mean': k, 'sd': np.abs(k * 2)})
shift_prior = ('Normal', {'mean': s, 'sd': np.abs(s * 2)})
priors = {
'amplitude': amp_prior,
'base': base_prior,
'kappa': kappa_prior,
'shift': shift_prior
}
if b_m:
priors['base_mod'] = base_prior
if a_m:
priors['amplitude_mod'] = amp_prior
if k_m:
priors['kappa_mod'] = kappa_prior
if s_m:
priors['shift_mod'] = shift_prior
modelspec[dsig_idx]['prior'] = priors
def before_and_after_scatter(rec, modelspec, idx, sig_name='pred',
compare='resp', smoothing_bins=False,
mod_name='Unknown', xlabel1=None, xlabel2=None,
ylabel1=None, ylabel2=None):
# HACK: shouldn't hardcode 'stim', might be named something else
# or not present at all. Need to figure out a better solution
# for special case of idx = 0
if idx == 0:
# Can't have anything before index 0, so use input stimulus
before = rec.copy()
before_sig = rec['stim']
before.name = '**stim'
else:
before = ms.evaluate(rec.copy(), modelspec, start=None, stop=idx)
before_sig = before[sig_name]
# now evaluate next module step
after = ms.evaluate(before.copy(), modelspec, start=idx, stop=idx+1)
after_sig = after[sig_name]
# compute correlation for pre-module before it's over-written
if before[sig_name].shape[0] == 1:
corr1 = nm.corrcoef(before, pred_name=sig_name, resp_name=compare)
corr2 = nm.corrcoef(after, pred_name=sig_name, resp_name=compare)
else:
corr1 = 0
corr2 = 0
log.warning('corr coef expects single-dim predictions')
compare_to = rec[compare]
title1 = '{} vs {} before {}'.format(sig_name, compare, mod_name)
title2 = '{} vs {} after {}'.format(sig_name, compare, mod_name)
# TODO: These are coming out the same, but that seems unlikely
text1 = "r = {0:.5f}".format(corr1)
text2 = "r = {0:.5f}".format(corr2)
modidx = find_module(mod_name, modelspec)
if modidx:
module = modelspec[modidx]
else:
module = None
fn1 = partial(plot_scatter, before_sig, compare_to, title=title1,
smoothing_bins=smoothing_bins, xlabel=xlabel1,
ylabel=ylabel1, text=text1, module=module)
fn2 = partial(plot_scatter, after_sig, compare_to, title=title2,
smoothing_bins=smoothing_bins, xlabel=xlabel2,
ylabel=ylabel2, text=text2)
return fn1, fn2
def init_logsig(rec, modelspec):
'''
Initialization of priors for logistic_sigmoid,
based on process described in methods of Rabinowitz et al. 2014.
'''
# preserve input modelspec
modelspec = copy.deepcopy(modelspec)
logsig_idx = find_module('logistic_sigmoid', modelspec)
if logsig_idx is None:
log.warning("No logsig module was found, can't initialize.")
return modelspec
stim = rec['stim'].as_continuous()
resp = rec['resp'].as_continuous()
# TODO: Maybe need a more sophisticated calculation for this?
# Paper isn't very clear on how they calculate "X-bar" and "Y-bar"
# They also mention that their stim-resp data is split up into 20
# bins, maybe averaged across trials or something?
mean_stim = np.nanmean(stim)
min_stim = np.nanmin(stim)
max_stim = np.nanmax(stim)
stim_range = max_stim - min_stim
min_resp = np.nanmin(resp)
max_resp = np.nanmax(resp)
resp_range = max_resp - min_resp
# Rather than setting a hard value for initial phi,
# set the prior distributions and let the fitter/analysis
# decide how to use it.
base = ('Exponential', {'beta': min_resp + 0.05 * (resp_range)})
amplitude = ('Exponential', {'beta': 2 * resp_range})
shift = ('Normal', {'mean': mean_stim, 'sd': stim_range})
kappa = ('Exponential', {'beta': stim_range / stim.shape[1]})
modelspec[logsig_idx]['prior'] = {
'base': base,
'amplitude': amplitude,
'shift': shift,
'kappa': kappa
}
log.info(
"logistic_sigmoid priors initialized to: "
"base: %s\namplitude: %s\nshift: %s\nkappa: %s\n",
*modelspec[logsig_idx]['prior'].values())
return modelspec
def remove_dsig_bounds(modelspec):
dsig_idx = find_module('dynamic_sigmoid', modelspec)
if dsig_idx is None:
log.warning("No dsig module was found, can't initialize.")
return modelspec
modelspec = copy.deepcopy(modelspec)
modelspec[dsig_idx]['bounds'].update({
'base': (1e-15, None),
'amplitude': (1e-15, None),
'shift': (None, None),
'kappa': (1e-15, None),
'amplitude_mod': (None, None),
'base_mod': (None, None),
'kappa_mod': (None, None),
'shift_mod': (None, None)
})
return modelspec
def _set_gc_phi(modelspec, base, amplitude, shift, kappa):
'''
Parameters given as differences, e.g. kappa = -0.5 means set
kappa_mod to be 0.5 less than kappa.
'''
dsig_idx = find_module('dynamic_sigmoid', modelspec)
p = modelspec[dsig_idx]['phi']
b = p['base']
a = p['amplitude']
s = p['shift']
k = p['kappa']
modelspec[dsig_idx]['phi'].update({
'base_mod': b + base, 'amplitude_mod': a + amplitude,
'shift_mod': s + shift, 'kappa_mod': k + kappa
})
return modelspec
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 stp_sigmoid_pred_matched(cellid, batch, modelname, LN, include_phi=True):
xfspec, ctx = xhelp.load_model_xform(cellid, batch, modelname)
ln_spec, ln_ctx = xhelp.load_model_xform(cellid, batch, LN)
modelspec = ctx['modelspec']
modelspec.recording = ctx['val']
val = ctx['val'].apply_mask()
ln_modelspec = ln_ctx['modelspec']
ln_modelspec.recording = ln_ctx['val']
ln_val = ln_ctx['val'].apply_mask()
pred_after_NL = val['pred'].as_continuous().flatten() # with stp
val_before_NL = ms.evaluate(ln_val, ln_modelspec, stop=-1)
pred_before_NL = val_before_NL['pred'].as_continuous().flatten() # no stp
stp_idx = find_module('stp', modelspec)
val_before_stp = ms.evaluate(val, modelspec, stop=stp_idx)
val_after_stp = ms.evaluate(val, modelspec, stop=stp_idx + 1)
pred_before_stp = val_before_stp['pred'].as_continuous().mean(
axis=0).flatten()
pred_after_stp = val_after_stp['pred'].as_continuous().mean(
axis=0).flatten()
stp_effect = (pred_after_stp - pred_before_stp) / (pred_after_stp +
pred_before_stp)
fig = plt.figure()
plasma = plt.get_cmap('plasma')
plt.scatter(pred_before_NL,
pred_after_NL,
c=stp_effect,
s=2,
alpha=0.75,
cmap=plasma)
plt.title(cellid)
plt.xlabel('pred in (no stp)')
plt.ylabel('pred out (with stp)')
if include_phi:
stp_phi = modelspec.phi[stp_idx]
phi_string = '\n'.join(
['%s: %.4E' % (k, v) for k, v in stp_phi.items()])
fig.text(0.775, 0.9, phi_string, va='top', ha='left')
plt.subplots_adjust(right=0.775, left=0.075)
plt.colorbar()
return fig
def pca_proj_layer(rec, modelspec, **ctx):
from nems.tf.cnnlink_new import fit_tf, fit_tf_init
weight_chan_idx = find_module("weight_channels", modelspec, find_all_matches=True)
w = weight_chan_idx[-1]
coefficients = modelspec.phi[w]['coefficients'].copy()
pcs_needed = int(np.ceil(coefficients.shape[1] / 2))
if 'state' in modelspec[w-1]['fn']:
w -= 1
try:
v = rec.meta['pc_weights'].T[:, :pcs_needed]
pc_rec = rec.copy()
log.info('Found %d sets of PC weights', pcs_needed)
except:
pc_rec = resp_to_pc(rec=rec, pc_count=pcs_needed, pc_source='all', overwrite_resp=False, **ctx)['rec']
v = pc_rec.meta['pc_weights'].T[:, :pcs_needed]
v = np.concatenate((v, -v), axis=1)
pc_modelspec = modelspec.copy()
for i in range(w,len(modelspec)):
pc_modelspec.pop_module()
d = pc_rec.signals['pca'].as_continuous()[:pcs_needed,:]
d = np.concatenate((d,-d), axis=0)
d = d[:coefficients.shape[1],:]
pc_rec['resp'] = pc_rec['resp']._modified_copy(data=d)
#_d = fit_tf_init(pc_modelspec, pc_rec, nl_init='skip', use_modelspec_init=True, epoch_name="")
_d = fit_tf_init(pc_modelspec, pc_rec, use_modelspec_init=True, epoch_name="")
pc_modelspec = _d['modelspec']
modelspec = modelspec.copy()
for i in range(w):
for k in modelspec.phi[i].keys():
modelspec.phi[i][k] = pc_modelspec.phi[i][k]
log.info('modelspec len: %d pc_modelspec len: %d', len(modelspec), len(pc_modelspec))
#pre = modelspec.phi[w]['coefficients'].std()
#modelspec.phi[w]['coefficients'] = v[:,:coefficients.shape[1]]
#post = modelspec.phi[w]['coefficients'].std()
#log.info('Pasted pc weights into N x R = %d x %d weight channels matrix %.3f -> %.3f', v.shape[0], v.shape[1], pre, post)
return {'modelspec': modelspec, 'pc_modelspec': pc_modelspec}
def state_gain_plot(modelspec, ax=None, colors=None, clim=None, title=None, **options):
state_idx = find_module('state', modelspec)
g = modelspec.phi_mean[state_idx]['g']
d = modelspec.phi_mean[state_idx]['d']
ge = modelspec.phi_sem[state_idx]['g']
de = modelspec.phi_sem[state_idx]['d']
MI = modelspec[0]['meta']['state_mod']
state_chans = modelspec[0]['meta']['state_chans']
if ax is not None:
plt.sca(ax)
else:
ax=plt.gca()
if d.shape[0] > 1:
opt={}
for cc in range(d.shape[1]):
if colors is not None:
opt = {'color': colors[cc]}
plt.plot(d[:,cc],'--', **opt)
plt.plot(g[:,cc], **opt)
else:
plt.errorbar(np.arange(len(d[0, :])), d[0, :], de[0, :], color='blue')
plt.errorbar(np.arange(len(g[0, :])), g[0, :], ge[0, :], color='red')
dz = np.abs(d[0, :] / de[0, :])
gz = np.abs(g[0, :] / ge[0, :])
for i in range(len(gz)):
if gz[i] > 2:
ax.text(i, g[0, i] + np.sign(g[0, i]) * ge[0, i], state_chans[i],
color='red', ha='center', fontsize=6)
elif dz[i] > 2:
ax.text(i, d[0,i]+np.sign(d[0,i])*de[0,i], state_chans[i],
color='blue', ha='center', fontsize=6)
#plt.plot(MI)
#plt.xticks(np.arange(len(state_chans)), state_chans, fontsize=6)
plt.legend(('baseline', 'gain'), frameon=False)
plt.plot(np.arange(len(state_chans)),np.zeros(len(state_chans)),'k--',
linewidth=0.5)
if title:
plt.title(title)
ax_remove_box(ax)
def single_state_mod_index(rec,
modelspec,
epoch='REFERENCE',
psth_name='pred',
state_sig='state',
state_chan='pupil'):
if type(state_chan) is list:
if len(state_chan) == 0:
state_chan = rec[state_sig].chans
mod_list = [
single_state_mod_index(rec,
modelspec,
epoch=epoch,
psth_name=psth_name,
state_sig=state_sig,
state_chan=s) for s in state_chan
]
return mod_list
sidx = find_module('state', modelspec)
if sidx is None:
raise ValueError("no state signal found")
modelspec = copy.deepcopy(modelspec)
state_chan_idx = rec[state_sig].chans.index(state_chan)
k = np.ones(rec[state_sig].shape[0], dtype=bool)
k[0] = False
k[state_chan_idx] = False
modelspec[sidx]['phi']['d'][:, k] = 0
modelspec[sidx]['phi']['g'][:, k] = 0
newrec = ms.evaluate(rec, modelspec)
return np.array(
state_mod_index(newrec,
epoch=epoch,
psth_name=psth_name,
state_sig=state_sig,
state_chan=state_chan))
def init_dsig(rec, modelspec):
'''
Initialization of priors for logistic_sigmoid,
based on process described in methods of Rabinowitz et al. 2014.
'''
dsig_idx = find_module('dynamic_sigmoid', modelspec)
if dsig_idx is None:
log.warning("No dsig module was found, can't initialize.")
return modelspec
modelspec = copy.deepcopy(modelspec)
rec = copy.deepcopy(rec)
if modelspec[dsig_idx]['fn_kwargs'].get('eq', '') in \
['dexp', 'd', 'double_exponential']:
modelspec = _init_double_exponential(rec, modelspec, dsig_idx)
else:
modelspec = _init_logistic_sigmoid(rec, modelspec, dsig_idx)
return modelspec
def contrast_kernel_heatmap(rec,
modelspec,
ax=None,
title=None,
idx=0,
channels=0,
xlabel='Lag (s)',
ylabel='Channel In',
**options):
ctk_idx = nu.find_module('contrast_kernel', modelspec)
phi = copy.deepcopy(modelspec[ctk_idx]['phi'])
fn_kwargs = copy.deepcopy(modelspec[ctk_idx]['fn_kwargs'])
fs = rec['stim'].fs
old = ('auto_copy' in fn_kwargs)
if old:
fn_kwargs['use_phi'] = True
# Remove duplicates from fn_kwargs (phi is more up to date)
# to avoid argument collisions
removals = []
for k in fn_kwargs:
if k in phi:
removals.append(k)
for k in removals:
fn_kwargs.pop(k)
strf, wc_coefs, fir_coefs = _get_ctk_coefficients(**fn_kwargs,
**phi,
fs=fs)
_strf_heatmap(strf,
wc_coefs,
fir_coefs,
xlabel=xlabel,
ylabel=ylabel,
ax=ax,
title=title)
return ax
def init_relsat(rec, modelspec):
modelspec = copy.deepcopy(modelspec)
target_i = find_module('saturated_rectifier', modelspec)
if target_i is None:
log.warning("No relsat module was found, can't initialize.")
return modelspec
if target_i == len(modelspec):
fit_portion = modelspec.modules
else:
fit_portion = modelspec.modules[:target_i]
# generate prediction from module preceeding dexp
#rec = ms.evaluate(rec, ms.ModelSpec(fit_portion)).apply_mask()
rec = ms.ModelSpec(fit_portion).evaluate(rec).apply_mask()
pred = rec['pred'].as_continuous().flatten()
resp = rec['resp'].as_continuous().flatten()
stdr = np.nanstd(resp)
base = np.min(resp)
amplitude = min(np.mean(resp) + stdr * 3, np.max(resp))
shift = np.mean(pred) - 1.5 * np.nanstd(pred)
kappa = 1
base_prior = ('Exponential', {'beta': base})
amplitude_prior = ('Exponential', {'beta': amplitude})
shift_prior = ('Normal', {'mean': shift, 'sd': shift})
kappa_prior = ('Exponential', {'beta': kappa})
modelspec['prior'] = {
'base': base_prior,
'amplitude': amplitude_prior,
'shift': shift_prior,
'kappa': kappa_prior
}
return modelspec
def strf_local_lin(rec, modelspec, cursor_time=20, channels=0, **options):
rec = rec.copy()
tbin = int(cursor_time * rec['resp'].fs)
chan_count = rec['stim'].shape[0]
firmod = find_module('fir', modelspec)
tbin_count = modelspec.phi[firmod]['coefficients'].shape[1] + 2
use_dstrf = True
if use_dstrf:
index = int(cursor_time * rec['resp'].fs)
strf = modelspec.get_dstrf(rec,
index=index,
width=20,
out_channel=channels)
else:
resp_chan = channels
d = rec['stim']._data.copy()
strf = np.zeros((chan_count, tbin_count))
_p1 = rec['pred']._data[resp_chan, tbin]
eps = np.nanstd(d) / 100
eps = 0.01
#print('eps: {}'.format(eps))
for c in range(chan_count):
#eps = np.std(d[c, :])/100
for t in range(tbin_count):
_d = d.copy()
_d[c, tbin - t] *= 1 + eps
rec['stim'] = rec['stim']._modified_copy(data=_d)
rec = modelspec.evaluate(rec)
_p2 = rec['pred']._data[resp_chan, tbin]
strf[c, t] = (_p2 - _p1) / eps
print('strf min: {} max: {}'.format(np.min(strf), np.max(strf)))
options['clim'] = np.array([-np.max(np.abs(strf)), np.max(np.abs(strf))])
plot_heatmap(strf, cmap=get_setting('FILTER_CMAP'), **options)
def init_dsig(rec, modelspec, nl_mode=2):
'''
Initialization of priors for logistic_sigmoid,
based on process described in methods of Rabinowitz et al. 2014.
'''
dsig_idx = find_module('dynamic_sigmoid', modelspec)
if dsig_idx is None:
log.warning("No dsig module was found, can't initialize.")
return modelspec
if modelspec[dsig_idx]['fn_kwargs'].get('eq', '') in \
['dexp', 'd', 'double_exponential']:
modelspec = _init_double_exponential(rec,
modelspec,
dsig_idx,
nl_mode=nl_mode)
elif modelspec[dsig_idx]['fn_kwargs'].get('eq', '') in \
['relsat', 'rs', 'saturated_rectifier']:
modelspec = init_relsat(rec, modelspec)
else:
modelspec = init_logsig(rec, modelspec)
return modelspec