from FiCurve import FICurve
from CellData import CellData
import matplotlib.pyplot as plt
from scipy.optimize import curve_fit
import os
import numpy as np
import functions as fu
class Adaption:
def __init__(self, cell_data: CellData, fi_curve: FICurve = None):
self.cell_data = cell_data
if fi_curve is None:
self.fi_curve = FICurve(cell_data)
else:
self.fi_curve = fi_curve
# [[a, tau_eff, c], [], [a, tau_eff, c], ...]
self.exponential_fit_vars = []
self.tau_real = []
self.fit_exponential()
self.calculate_tau_from_tau_eff()
def fit_exponential(self, length_of_fit=0.1):
mean_frequencies = self.cell_data.get_mean_isi_frequencies()
time_axes = self.cell_data.get_time_axes_mean_frequencies()
for i in range(len(mean_frequencies)):
start_idx = self.__find_start_idx_for_exponential_fit(i)
if start_idx == -1:
print("start index negative")
self.exponential_fit_vars.append([])
continue
# shorten length of fit to stay in stimulus region if given length is too long
sampling_interval = self.cell_data.get_sampling_interval()
used_length_of_fit = length_of_fit
if (start_idx * sampling_interval) - self.cell_data.get_delay() + length_of_fit > self.cell_data.get_stimulus_end():
print(start_idx * sampling_interval, "start - end", start_idx * sampling_interval + length_of_fit)
print("Shortened length of fit to keep it in the stimulus region!")
used_length_of_fit = self.cell_data.get_stimulus_end() - (start_idx * sampling_interval)
end_idx = start_idx + int(used_length_of_fit/sampling_interval)
y_values = mean_frequencies[i][start_idx:end_idx+1]
x_values = time_axes[i][start_idx:end_idx+1]
tau = self.__approximate_tau_for_exponential_fit(x_values, y_values, i)
# start the actual fit:
try:
p0 = (self.fi_curve.f_zeros[i], tau, self.fi_curve.f_infinities[i])
popt, pcov = curve_fit(fu.exponential_function, x_values, y_values,
p0=p0, maxfev=10000, bounds=([-np.inf, 0, -np.inf], [np.inf, np.inf, np.inf]))
except RuntimeError:
print("RuntimeError happened in fit_exponential.")
self.exponential_fit_vars.append([])
continue
# Obviously a bad fit - time constant, expected in range 3-10ms, has value over 1 second or is negative
if abs(popt[1] > 1) or popt[1] < 0:
print("detected an obviously bad fit")
self.exponential_fit_vars.append([])
else:
self.exponential_fit_vars.append(popt)
def __approximate_tau_for_exponential_fit(self, x_values, y_values, mean_freq_idx):
if self.fi_curve.f_infinities[mean_freq_idx] < self.fi_curve.f_baselines[mean_freq_idx] * 0.95:
test_val = [y > 0.65 * self.fi_curve.f_infinities[mean_freq_idx] for y in y_values]
else:
test_val = [y < 0.65 * self.fi_curve.f_zeros[mean_freq_idx] for y in y_values]
try:
idx = test_val.index(True)
if idx == 0:
idx = 1
tau = x_values[idx] - x_values[0]
except ValueError:
tau = x_values[-1] - x_values[0]
return tau
def __find_start_idx_for_exponential_fit(self, mean_freq_idx):
time_axes = self.cell_data.get_time_axes_mean_frequencies()[mean_freq_idx]
stimulus_start_idx = int((self.cell_data.get_stimulus_start() + time_axes[0]) / self.cell_data.get_sampling_interval())
if self.fi_curve.f_infinities[mean_freq_idx] > self.fi_curve.f_baselines[mean_freq_idx] * 1.1:
# start setting starting variables for the fit
# search for the start_index by searching for the max
j = 0
while True:
try:
if self.cell_data.get_mean_isi_frequencies()[mean_freq_idx][stimulus_start_idx + j] == self.fi_curve.f_zeros[mean_freq_idx]:
start_idx = stimulus_start_idx + j
break
except IndexError as e:
return -1
j += 1
elif self.fi_curve.f_infinities[mean_freq_idx] < self.fi_curve.f_baselines[mean_freq_idx] * 0.9:
# start setting starting variables for the fit
# search for start by finding the end of the minimum
found_min = False
j = int(0.05 / self.cell_data.get_sampling_interval())
nothing_to_fit = False
while True:
if not found_min:
if self.cell_data.get_mean_isi_frequencies()[mean_freq_idx][stimulus_start_idx + j] == self.fi_curve.f_zeros[mean_freq_idx]:
found_min = True
else:
if self.cell_data.get_mean_isi_frequencies()[mean_freq_idx][stimulus_start_idx + j + 1] > self.fi_curve.f_zeros[mean_freq_idx]:
start_idx = stimulus_start_idx + j
break
if j > 0.1 / self.cell_data.get_sampling_interval():
# no rise in freq until to close to the end of the stimulus (to little place to fit)
return -1
j += 1
if nothing_to_fit:
return -1
else:
# there is nothing to fit to:
return -1
return start_idx
def calculate_tau_from_tau_eff(self):
tau_effs = []
for i in range(len(self.exponential_fit_vars)):
if len(self.exponential_fit_vars[i]) == 0:
continue
tau_effs.append(self.exponential_fit_vars[i][1])
f_infinity_slope = self.fi_curve.get_f_infinity_slope()
# --- old way to calculate with the fi slope at middle of the fi curve
# fi_curve_slope = self.fi_curve.get_fi_curve_slope_of_straight()
# self.tau_real = np.median(tau_effs) * (fi_curve_slope / f_infinity_slope)
# print("fi_slope to f_inf slope:", fi_curve_slope/f_infinity_slope)
# print("fi_slope:", fi_curve_slope, "f_inf slope:", f_infinity_slope)
# print("current tau: {:.1f}ms".format(np.median(tau_effs) * (fi_curve_slope / f_infinity_slope) * 1000))
# new way to calculate with the fi curve slope at the intersection point of it and the f_inf line
factor = self.fi_curve.get_fi_curve_slope_at_f_zero_intersection() / f_infinity_slope
self.tau_real = np.median(tau_effs) * factor
print("###### tau: {:.1f}ms".format(self.tau_real*1000), "other f_0 slope:", self.fi_curve.get_fi_curve_slope_at_f_zero_intersection())
def get_tau_real(self):
return np.median(self.tau_real)
def get_tau_effs(self):
return [ex_vars[1] for ex_vars in self.exponential_fit_vars if ex_vars != []]
def plot_exponential_fits(self, save_path: str = None, indices: list = None, delete_previous: bool = False):
if delete_previous:
for val in self.cell_data.get_fi_contrasts():
prev_path = save_path + "mean_freq_exp_fit_contrast:" + str(round(val, 3)) + ".png"
if os.path.exists(prev_path):
os.remove(prev_path)
for i in range(len(self.cell_data.get_fi_contrasts())):
if indices is not None and i not in indices:
continue
if self.exponential_fit_vars[i] == []:
print("no fit vars for index!")
continue
plt.plot(self.cell_data.get_time_axes_mean_frequencies()[i], self.cell_data.get_mean_isi_frequencies()[i])
vars = self.exponential_fit_vars[i]
fit_x = np.arange(0, 0.4, self.cell_data.get_sampling_interval())
plt.plot(fit_x, [fu.exponential_function(x, vars[0], vars[1], vars[2]) for x in fit_x])
plt.ylim([0, max(self.fi_curve.f_zeros[i], self.fi_curve.f_baselines[i])*1.1])
plt.xlabel("Time [s]")
plt.ylabel("Frequency [Hz]")
if save_path is None:
plt.show()
else:
plt.savefig(save_path + "mean_freq_exp_fit_contrast:" + str(round(self.cell_data.get_fi_contrasts()[i], 3)) + ".png")
plt.close()