from models.LIFACnoise import LifacNoiseModel
from Baseline import BaselineModel
from FiCurve import FICurveModel
import numpy as np
import matplotlib.pyplot as plt
import copy
import os
SEARCH_WIDTH = 3
SEARCH_PRECISION = 40
CONTRASTS = np.arange(-0.4, 0.45, 0.1)
def main():
test_effect_of_two_variables()
quit()
model_parameters1 = {'threshold': 1,
'step_size': 5e-05,
'a_zero': 2,
'delta_a': 0.2032269898801589,
'mem_tau': 0.011314027210564803,
'noise_strength': 0.056724809998220195,
'v_zero': 0,
'v_base': 0,
'tau_a': 0.05958195972016753,
'input_scaling': 119.81500448274554,
'dend_tau': 0.0027746086464721723,
'v_offset': -24.21875,
'refractory_period': 0.0006}
model_parameters2 = {'v_offset': -15.234375, 'input_scaling': 64.94152780134829, 'step_size': 5e-05, 'a_zero': 2,
'threshold': 1, 'v_base': 0, 'delta_a': 0.04763179657857666, 'tau_a': 0.07891848949732623,
'mem_tau': 0.004828473985707999, 'noise_strength': 0.017132801387559883,
'v_zero': 0, 'dend_tau': 0.0015230454266819539, 'refractory_period': 0.0006}
parameters_to_test = ["input_scaling", "refractory_period", "dend_tau", "mem_tau", "noise_strength", "v_offset", "delta_a", "tau_a"]
# parameters_to_test = ["refractory_period", "input_scaling"]
effect_data = []
for p in parameters_to_test:
print("Working on parameter " + p)
effect_data.append(test_parameter_effect(model_parameters2, p, 600))
plot_effects(effect_data, "./figures/variable_effect/")
def test_effect_of_two_variables():
eod_freqs = np.arange(100, 1001, 20)
ref_periods = np.arange(0, 0.00201, 0.0001)
variables = ("bf", "vs", "sc", "cv", "burst", "f_inf_s", "f_zero_s")
colorbar_labels = ("Frequency in Hz", "Vector strength", "serial correlation lag=1", "Coefficient of Variation",
"Burstiness", "f_inf slope", "f_zero slope")
# base eod frequency would be 771!
base_parameters = {'step_size': 5e-05, 'mem_tau': 0.0076958612706114595, 'v_base': 0, 'v_zero': 0, 'threshold': 1,
'v_offset': -37.5, 'input_scaling': 181.40702315746051, 'delta_a': 0.333391796423963,
'tau_a': 0.17301586167067445, 'a_zero': 2, 'noise_strength': 0.017424670423939775,
'dend_tau': 0.0037179224836952356, 'refractory_period': 0.0010602702699897444}
# base eod frequency would be 657!
base_parameters = {'refractory_period': 0.0008347981797599925, 'v_base': 0, 'v_zero': 0, 'a_zero': 2,
'step_size': 5e-05, 'delta_a': 0.10570085698152036, 'threshold': 1,
'input_scaling': 85.7818875779873, 'mem_tau': 0.01094261953657057, 'tau_a': 0.07741757133763925,
'v_offset': -10.15625, 'noise_strength': 0.03080655781041302, 'dend_tau': 0.0013624430015225777}
effects = []
for eod_freq in eod_freqs:
effects_with_const_eod_freq = []
for ref_period in ref_periods:
model_parameters = copy.deepcopy(base_parameters)
model_parameters["refractory_period"] = ref_period
effects_with_const_eod_freq.append(test_model(model_parameters, eod_freq))
effects.append(effects_with_const_eod_freq)
if not os.path.exists("./figures/eod_and_ref_period_effect/"):
os.makedirs("./figures/eod_and_ref_period_effect/")
for x, variable in enumerate(variables):
matrix = np.zeros((len(eod_freqs), len(ref_periods)))
for i in range(len(eod_freqs)):
for j in range(len(ref_periods)):
matrix[i, j] = effects[i][j][variable]
fig, axes = plt.subplots(nrows=1, ncols=1, figsize=(6, 6))
im = axes.imshow(matrix)
cbar = axes.figure.colorbar(im, ax=axes)
cbar.ax.set_ylabel(colorbar_labels[x], rotation=-90, va="bottom")
axes.set_title(variable)
axes.set_xlabel("Refractory periods in ms")
axes.set_ylabel("EOD frequency in Hz")
axes.set_xticks(np.arange(len(ref_periods)))
axes.set_yticks(np.arange(len(eod_freqs)))
# ... and label them with the respective list entries
axes.set_xticklabels(["{:.2f}".format(r*1000) for r in ref_periods])
axes.set_yticklabels(eod_freqs)
plt.setp(axes.get_xticklabels(), rotation=45, ha="right",
rotation_mode="anchor")
axes.set_ylabel("Eod frequencies")
plt.tight_layout()
plt.savefig("./figures/eod_and_ref_period_effect/" + variable + ".png")
plt.close()
def test_model(model_parameters, eod_freq):
model = LifacNoiseModel(model_parameters)
print(model.get_parameters())
fi_curve = FICurveModel(model, CONTRASTS, eod_freq, trials=10)
f_inf_s = fi_curve.get_f_inf_slope()
f_inf_v = fi_curve.get_f_inf_frequencies()
f_zero_s = fi_curve.get_f_zero_fit_slope_at_stimulus_value(0.1)
f_zero_v = fi_curve.get_f_zero_frequencies()
baseline = BaselineModel(model, eod_freq, trials=3)
bf = baseline.get_baseline_frequency()
vs = baseline.get_vector_strength()
sc = baseline.get_serial_correlation(1)[0]
cv = baseline.get_coefficient_of_variation()
burst = baseline.get_burstiness()
return {"f_inf_s": f_inf_s, "f_inf_v": f_inf_v, "f_zero_s": f_zero_s, "f_zero_v": f_zero_v,
"bf": bf, "vs": vs, "sc": sc, "cv": cv, "burst": burst}
def test_parameter_effect(model_parameters, test_parameter, eod_freq):
model_parameters = copy.deepcopy(model_parameters)
start_value = model_parameters[test_parameter]
start = start_value*(1/SEARCH_WIDTH)
end = start_value*SEARCH_WIDTH
step = (end - start) / SEARCH_PRECISION
values = np.arange(start, end+step, step)
bf = []
vs = []
sc = []
cv = []
burst = []
f_inf_s = []
f_inf_v = []
f_zero_s = []
f_zero_v = []
broken_i = []
for i in range(len(values)):
model_parameters[test_parameter] = values[i]
model = LifacNoiseModel(model_parameters)
fi_curve = FICurveModel(model, CONTRASTS, eod_freq, trials=10)
f_inf_s.append(fi_curve.get_f_inf_slope())
f_inf_v.append(fi_curve.get_f_inf_frequencies())
f_zero_s.append(fi_curve.get_f_zero_fit_slope_at_stimulus_value(0.1))
f_zero_v.append(fi_curve.get_f_zero_frequencies())
if not os.path.exists("./figures/f_point_detection/"):
os.makedirs("./figures/f_point_detection/")
detection_save_path = "./figures/f_point_detection/{}_{:.4f}/".format(test_parameter, values[i])
if not os.path.exists(detection_save_path):
os.makedirs(detection_save_path)
fi_curve.plot_f_point_detections(detection_save_path)
baseline = BaselineModel(model, eod_freq, trials=3)
bf.append(baseline.get_baseline_frequency())
vs.append(baseline.get_vector_strength())
sc.append(baseline.get_serial_correlation(2))
cv.append(baseline.get_coefficient_of_variation())
burst.append(baseline.get_burstiness())
values = list(values)
if len(broken_i) > 0:
broken_i = sorted(broken_i, reverse=True)
for i in broken_i:
del values[i]
return ParameterEffectData(values, test_parameter, bf, vs, sc, cv, burst, f_inf_s, f_inf_v, f_zero_s, f_zero_v)
# plot_effects(values, test_parameter, bf, vs, sc, cv, f_inf_s, f_inf_v, f_zero_s, f_zero_v)
def plot_effects(par_effect_data_list, save_path=None):
names = ("bf", "vs", "sc", "cv", "burstiness", "f_inf_s", "f_inf_v", "f_zero_s", "f_zero_v")
fig, axes = plt.subplots(len(names), len(par_effect_data_list), figsize=(4*len(par_effect_data_list), 4*len(names)), sharex="col")
for j in range(len(par_effect_data_list)):
ped = par_effect_data_list[j]
ranges = ((0, max(ped.get_data("bf")) * 1.1), (0, 1), (-1, 1), (0, 1), (0, 1),
(0, max(ped.get_data("f_inf_s")) * 1.1), (0, 800),
(0, max(ped.get_data("f_zero_s")) * 1.1), (0, 10000))
values = ped.values
for i in range(len(names)):
y_data = ped.get_data(names[i])
axes[i, j].plot(values, y_data)
if names[i] == "f_zero_v":
axes[i, j].set_yscale('log')
axes[i, j].set_ylim(ranges[i])
else:
axes[i, j].set_ylim(ranges[i])
if j == 0:
axes[i, j].set_ylabel(names[i])
if i == 0:
axes[i, j].set_title(ped.test_parameter)
plt.tight_layout()
if save_path is not None:
plt.savefig(save_path + "variable_effect_master_plot.png")
else:
plt.show()
plt.close()
class ParameterEffectData:
data_names = ("bf", "vs", "sc", "cv", "burstiness","f_inf_s", "f_inf_v" "f_zero_s", "f_zero_v")
def __init__(self, values, test_parameter, bf, vs, sc, cv, burstiness, f_inf_s, f_inf_v, f_zero_s, f_zero_v):
self.values = values
self.test_parameter = test_parameter
self.bf = bf
self.vs = vs
self.sc = sc
self.cv = cv
self.f_inf_s = f_inf_s
self.f_inf_v = f_inf_v
self.f_zero_s = f_zero_s
self.f_zero_v = f_zero_v
self.burstiness = burstiness
def get_data(self, name):
if name == "bf":
return self.bf
elif name == "vs":
return self.vs
elif name == "sc":
return self.sc
elif name == "cv":
return self.cv
elif name == "f_inf_s":
return self.f_inf_s
elif name == "f_inf_v":
return self.f_inf_v
elif name == "f_zero_s":
return self.f_zero_s
elif name == "f_zero_v":
return self.f_zero_v
elif name == "burstiness":
return self.burstiness
else:
raise ValueError("Unknown attribute name!")
if __name__ == '__main__':
main()