import numpy as np
from sympy.abc import x as symbolic_x
from sympy.abc import y as symbolic_y
from six import string_types
from bmtk.simulator.filternet.filters import TemporalFilterCosineBump, GaussianSpatialFilter, SpatioTemporalFilter, \
WaveletFilter, SpectroTemporalFilter, GaborFilter
from bmtk.simulator.filternet.cell_models import TwoSubfieldLinearCell, OnUnit, OffUnit, LGNOnOffCell
from bmtk.simulator.filternet.transfer_functions import ScalarTransferFunction, MultiTransferFunction
from bmtk.simulator.filternet.utils import get_data_metrics_for_each_subclass, get_tcross_from_temporal_kernel
from bmtk.simulator.filternet.pyfunction_cache import py_modules
[docs]
def create_two_sub_cell(dom_lf, non_dom_lf, dom_spont, non_dom_spont, onoff_axis_angle, subfield_separation,
dom_location):
dsp = str(dom_spont)
ndsp = str(non_dom_spont)
two_sub_transfer_fn = MultiTransferFunction((symbolic_x, symbolic_y),
'Heaviside(x+'+dsp+')*(x+'+dsp+')+Heaviside(y+'+ndsp+')*(y+'+ndsp+')')
two_sub_cell = TwoSubfieldLinearCell(dom_lf, non_dom_lf, subfield_separation=subfield_separation,
onoff_axis_angle=onoff_axis_angle, dominant_subfield_location=dom_location,
transfer_function=two_sub_transfer_fn)
return two_sub_cell
[docs]
def createOneUnitOfTwoSubunitFilter(weights, kpeaks, delays, ttp_exp):
delays = np.array(delays)
filt = TemporalFilterCosineBump(weights, kpeaks, delays)
tcross_ind = get_tcross_from_temporal_kernel(filt.get_kernel(threshold=-1.0).kernel)
filt_sum = filt.get_kernel(threshold=-1.0).kernel[:tcross_ind].sum()
# Calculate delay offset needed to match response latency with data and rebuild temporal filter
del_offset = ttp_exp - tcross_ind
if del_offset >= 0:
delays_updated = delays + del_offset
filt_new = TemporalFilterCosineBump(weights, kpeaks, delays_updated)
else:
raise Exception('del_offset < 0')
return filt_new, filt_sum
[docs]
def get_tf_params(node, dynamics_params, non_dom_props=False):
if not non_dom_props:
weights = node.weights if node.weights is not None else dynamics_params['opt_wts']
kpeaks = node.kpeaks if node.kpeaks is not None else dynamics_params['opt_kpeaks']
delays = node.delays if node.delays is not None else dynamics_params['opt_delays']
else:
dp = dynamics_params or {}
weights = node.weights_non_dom if node.weights_non_dom is not None else dp.get('opt_wts', None)
kpeaks = node.kpeaks_non_dom if node.kpeaks_non_dom is not None else dp.get('opt_kpeaks', None)
delays = node.delays_non_dom if node.delays_non_dom is not None else dp.get('opt_delays', None)
if node.predefined_jitter:
jitter_fnc = lambda a: np.array([np.random.uniform(x*node.jitter[0], x*node.jitter[1]) for x in a])
weights = jitter_fnc(weights) if weights is not None else weights
kpeaks = jitter_fnc(kpeaks) if kpeaks is not None else kpeaks
delays = jitter_fnc(delays) if delays is not None else delays
return weights, kpeaks, delays
[docs]
def get_sigma(node, dynamics_params):
if 'spatial_size' in node:
sigma = node['spatial_size']
elif 'sigma' in node:
sigma = node['sigma']
elif 'spatial_size' in dynamics_params:
sigma = dynamics_params['spatial_size']
elif 'sigma' in dynamics_params:
sigma = dynamics_params['spatial_size']
else:
# TODO: Raise warning
sigma = (1.0, 1.0)
if np.isscalar(sigma):
sigma = (sigma, sigma) # convert from degree to SD
return sigma[0]/3.0, sigma[1]/3.0
[docs]
def get_wavelet_params(node, dynamics_params):
t_mod_freq = node.t_mod_freq if node.t_mod_freq is not None else dynamics_params['t_mod_freq']
sp_mod_freq = node.sp_mod_freq if node.sp_mod_freq is not None else dynamics_params['sp_mod_freq']
if (t_mod_freq < 0) or (sp_mod_freq < 0):
raise Exception("Temporal modulation frequency (t_mod_freq) and spectral modulation frequency "
"(sp_mod_freq) must be non-negative.")
Lambda = 1/np.linalg.norm([t_mod_freq, sp_mod_freq]) # Wavelength of oscillatory component
sigma_f = node.sigma_f if node.sigma_f is not None else dynamics_params['sigma_f']
b_t = node.b_t if node.b_t is not None else dynamics_params['b_t']
order_t = node.order_t if node.order_t is not None else dynamics_params['order_t']
amplitude = node.amplitude if node.amplitude is not None else dynamics_params['amplitude']
if t_mod_freq != 0:
theta = np.arctan(sp_mod_freq / t_mod_freq)
else:
theta = np.pi / 2
psi = node.psi if node.psi is not None else dynamics_params['psi']
if isinstance(psi, string_types):
psi = eval(psi.replace('pi', 'np.pi'))
delay = node.delay if node.delay is not None else dynamics_params['delay']
direction = node.direction if node.direction is not None else dynamics_params['direction']
if direction == 'up':
direction = 1
elif direction == 'down':
#dynamics_params['direction'] = -1
direction = -1
elif direction in [-1,0,1]:
pass
else:
raise Exception("'Direction' filter parameter must be 'up' (or 1) for upward frequency modulation, "
" or 'down' (or -1) for downward modulation, or 0 if not applicable.")
return Lambda, sigma_f, b_t, order_t, theta, psi, delay, amplitude, direction
[docs]
def default_cell_loader(node, template_name, dynamics_params):
"""
:param node:
:param template_name:
:param dynamics_params:
:return:
"""
if template_name is None or template_name[0] == 'lgnmodel':
# Create the spatial filter
origin = (0.0, 0.0)
translate = (node['x'], node['y'])
sigma = get_sigma(node, dynamics_params)
if 'spatial_rotation' in node:
rotation = node['spatial_rotation']
else:
rotation = 0.0
t_weights, t_kpeaks, t_delays = get_tf_params(node, dynamics_params)
if template_name:
model_name = template_name[1]
else:
model_name = node['pop_name']
if model_name in ['sONsOFF_001', 'sONsOFF']:
# sON temporal filter
t_weights_nd, t_kpeaks_nd, t_delays_nd = get_tf_params(node, node.non_dom_params, non_dom_props=True)
sON_filt_new, sON_sum = createOneUnitOfTwoSubunitFilter(t_weights_nd, t_kpeaks_nd, t_delays_nd, 121.0)
sOFF_filt_new, sOFF_sum = createOneUnitOfTwoSubunitFilter(t_weights, t_kpeaks, t_delays, 115.0)
amp_on = 1.0 # set the non-dominant subunit amplitude to unity
spont = 4.0
max_roff = 35.0
max_ron = 21.0
amp_off = -(max_roff/max_ron)*(sON_sum/sOFF_sum)*amp_on - (spont*(max_roff - max_ron))/(max_ron*sOFF_sum)
# Create sON subunit:
spatial_filter_on = GaussianSpatialFilter(translate=translate, sigma=sigma, origin=origin, rotation=rotation)
linear_filter_son = SpatioTemporalFilter(spatial_filter_on, sON_filt_new, amplitude=amp_on)
# Create sOFF subunit:
spatial_filter_off = GaussianSpatialFilter(translate=translate, sigma=sigma, origin=origin, rotation=rotation)
linear_filter_soff = SpatioTemporalFilter(spatial_filter_off, sOFF_filt_new, amplitude=amp_off)
sf_sep = node.sf_sep
if node.predefined_jitter:
sf_sep = np.random.uniform(node.jitter[0]*sf_sep, node.jitter[1]*sf_sep)
sep_ss_onoff_cell = create_two_sub_cell(linear_filter_soff, linear_filter_son, 0.5 * spont, 0.5 * spont,
node.tuning_angle, sf_sep, translate)
cell = sep_ss_onoff_cell
elif model_name in ['sONtOFF_001', 'sONtOFF']:
t_weights_nd, t_kpeaks_nd, t_delays_nd = get_tf_params(node, node.non_dom_params, non_dom_props=True)
sON_filt_new, sON_sum = createOneUnitOfTwoSubunitFilter(t_weights_nd, t_kpeaks_nd, t_delays_nd, 93.5)
tOFF_filt_new, tOFF_sum = createOneUnitOfTwoSubunitFilter(t_weights, t_kpeaks, t_delays, 64.8) # 64.8
amp_on = 1.0 # set the non-dominant subunit amplitude to unity
spont = 5.5
max_roff = 46.0
max_ron = 31.0
amp_off = -0.7*(max_roff/max_ron)*(sON_sum/tOFF_sum)*amp_on - (spont*(max_roff - max_ron))/(max_ron*tOFF_sum)
# Create sON subunit:
spatial_filter_on = GaussianSpatialFilter(translate=translate, sigma=sigma, origin=origin, rotation=rotation)
linear_filter_son = SpatioTemporalFilter(spatial_filter_on, sON_filt_new, amplitude=amp_on)
# Create tOFF subunit:
spatial_filter_off = GaussianSpatialFilter(translate=translate, sigma=sigma, origin=origin, rotation=rotation)
linear_filter_toff = SpatioTemporalFilter(spatial_filter_off, tOFF_filt_new, amplitude=amp_off)
sf_sep = node.sf_sep
if node.predefined_jitter:
sf_sep = np.random.uniform(node.jitter[0]*sf_sep, node.jitter[1]*sf_sep)
sep_ts_onoff_cell = create_two_sub_cell(linear_filter_toff, linear_filter_son, 0.5 * spont, 0.5 * spont,
node.tuning_angle, sf_sep, translate)
cell = sep_ts_onoff_cell
elif model_name == 'LGNOnOFFCell':
wts = [node['weight_dom_0'], node['weight_dom_1']]
kpeaks = [node['kpeaks_dom_0'], node['kpeaks_dom_1']]
delays = [node['delay_dom_0'], node['delay_dom_1']]
# transfer_function = ScalarTransferFunction('s')
temporal_filter = TemporalFilterCosineBump(wts, kpeaks, delays)
spatial_filter_on = GaussianSpatialFilter(sigma=node['sigma_on'], origin=origin, translate=translate)
on_linear_filter = SpatioTemporalFilter(spatial_filter_on, temporal_filter, amplitude=20)
spatial_filter_off = GaussianSpatialFilter(sigma=node['sigma_off'], origin=origin, translate=translate)
off_linear_filter = SpatioTemporalFilter(spatial_filter_off, temporal_filter, amplitude=-20)
cell = LGNOnOffCell(on_linear_filter, off_linear_filter)
else:
type_split = model_name.split('_')
if len(type_split) == 1:
cell_type = model_name
tf_str = 'TF8'
else:
cell_type, tf_str = type_split[0], type_split[1]
# Get spontaneous firing rate, either from the cell property of calculate from experimental data
if 'spont_fr' in node:
spont_fr = node['spont_fr']
else:
exp_prs_dict = get_data_metrics_for_each_subclass(cell_type)
subclass_prs_dict = exp_prs_dict[tf_str]
spont_fr = subclass_prs_dict['spont_exp'][0]
# Get filters
transfer_function = ScalarTransferFunction('Heaviside(s+{})*(s+{})'.format(spont_fr, spont_fr))
temporal_filter = TemporalFilterCosineBump(t_weights, t_kpeaks, t_delays)
spatial_filter = GaussianSpatialFilter(translate=translate, sigma=sigma, origin=origin, rotation=rotation)
if cell_type.find('ON') >= 0:
amplitude = 1.0
linear_filter = SpatioTemporalFilter(spatial_filter, temporal_filter, amplitude=amplitude)
cell = OnUnit(linear_filter, transfer_function)
elif cell_type.find('OFF') >= 0:
amplitude = -1.0
linear_filter = SpatioTemporalFilter(spatial_filter, temporal_filter, amplitude=amplitude)
cell = OffUnit(linear_filter, transfer_function)
elif template_name[0] == 'audmodel':
# Currently tying y to center freq
translate = (node['y'])
Lambda, sigma_f, b_t, order_t, theta, psi, delay, amplitude, direction = get_wavelet_params(node, dynamics_params)
spectrotemporal_filter = WaveletFilter(translate, sigma_f, b_t, order_t, theta, Lambda, psi, delay,
amplitude, direction)
#spectrotemporal_filter = GaborFilter(translate, sigma_f, Lambda/np.cos(theta)/4, theta, Lambda, psi, amplitude,
# direction)
if template_name:
model_name = template_name[1]
else:
model_name = node['pop_name']
if model_name == 'AUD_filt':
# Create the spectro-temporal filter
pass
# transfer_function = ScalarTransferFunction('Heaviside(s+{})*(s+{})'.format(spont_fr, spont_fr))
cell_type = model_name
# Get spontaneous firing rate, either from the cell property of calculate from experimental data
if 'spont_fr' in node:
spont_fr = node['spont_fr']
else:
'''
exp_prs_dict = get_data_metrics_for_each_subclass(cell_type)
subclass_prs_dict = exp_prs_dict[tf_str]
spont_fr = subclass_prs_dict['spont_exp'][0]
'''
spont_fr = 2
transfer_function = ScalarTransferFunction('Heaviside(s+{})*(s+{})'.format(spont_fr, spont_fr))
amplitude = 1.0
linear_filter = SpectroTemporalFilter(spectrotemporal_filter, amplitude=amplitude)
cell = OnUnit(linear_filter, transfer_function)
else:
pass
return cell
py_modules.add_cell_processor('default', default_cell_loader, overwrite=False)
py_modules.add_cell_processor('preset_params', default_cell_loader, overwrite=False)
