326 lines
9.3 KiB
Python
326 lines
9.3 KiB
Python
import glob
|
|
import pathlib
|
|
import numpy as np
|
|
import matplotlib.pyplot as plt
|
|
import rlxnix as rlx
|
|
from IPython import embed
|
|
from scipy.signal import welch
|
|
|
|
def AM(EODf, stimulus):
|
|
"""
|
|
Calculates the Amplitude Modulation and Nyquist frequency
|
|
|
|
Parameters
|
|
----------
|
|
EODf : float or int
|
|
The current EODf.
|
|
stimulus : float or int
|
|
The absolute frequency of the stimulus.
|
|
|
|
Returns
|
|
-------
|
|
AM : float
|
|
The amplitude modulation resulting from the stimulus.
|
|
nyquist : float
|
|
The maximum frequency possible to resolve with the EODf.
|
|
|
|
"""
|
|
nyquist = EODf * 0.5
|
|
AM = np.mod(stimulus, nyquist)
|
|
return AM, nyquist
|
|
|
|
def binary_spikes(spike_times, duration, dt):
|
|
"""
|
|
Converts the spike times to a binary representations
|
|
|
|
Parameters
|
|
----------
|
|
spike_times : np.array
|
|
The spike times.
|
|
duration : float
|
|
The trial duration:
|
|
dt : float
|
|
The temporal resolution.
|
|
|
|
Returns
|
|
-------
|
|
binary : np.array
|
|
The binary representation of the spike train.
|
|
|
|
"""
|
|
binary = np.zeros(int(np.round(duration / dt))) #create the binary array with the same length as potential
|
|
|
|
spike_indices = np.asarray(np.round(spike_times / dt), dtype = int) # get the indices
|
|
binary[spike_indices] = 1 # put the indices into binary
|
|
return binary
|
|
|
|
def extract_stim_data(stimulus):
|
|
'''
|
|
extracts all necessary metadata for each stimulus
|
|
|
|
Parameters
|
|
----------
|
|
stimulus : Stimulus object or rlxnix.base.repro module
|
|
The stimulus from which the data is needed.
|
|
|
|
Returns
|
|
-------
|
|
amplitude : float
|
|
The relative signal amplitude in percent.
|
|
df : float
|
|
Distance of the stimulus to the current EODf.
|
|
eodf : float
|
|
Current EODf.
|
|
stim_freq : float
|
|
The total stimulus frequency (EODF+df).
|
|
amp_mod : float
|
|
The current amplitude modulation.
|
|
ny_freq : float
|
|
The current nyquist frequency.
|
|
|
|
'''
|
|
# extract metadata
|
|
# the stim.name adjusts the first key as it changes with every stimulus
|
|
amplitude = stimulus.metadata[stimulus.name]['Contrast'][0][0]
|
|
df = stimulus.metadata[stimulus.name]['DeltaF'][0][0]
|
|
eodf = round(stimulus.metadata[stimulus.name]['EODf'][0][0])
|
|
stim_freq = round(stimulus.metadata[stimulus.name]['Frequency'][0][0])
|
|
# calculates the amplitude modulation
|
|
amp_mod, ny_freq = AM(eodf, stim_freq)
|
|
return amplitude, df, eodf, stim_freq, amp_mod, ny_freq
|
|
|
|
def firing_rate(binary_spikes, dt = 0.000025, box_width = 0.01):
|
|
'''
|
|
Calculates the firing rate from binary spikes
|
|
|
|
Parameters
|
|
----------
|
|
binary_spikes : np.array
|
|
The binary representation of the spike train.
|
|
dt : float, optional
|
|
Time difference between two datapoints. The default is 0.000025.
|
|
box_width : float, optional
|
|
Time window on which the rate should be computed on. The default is 0.01.
|
|
|
|
Returns
|
|
-------
|
|
rate : np.array
|
|
Array of firing rates.
|
|
|
|
'''
|
|
box = np.ones(int(box_width // dt))
|
|
box /= np.sum(box) * dt # normalisierung des box kernels to an integral of one
|
|
rate = np.convolve(binary_spikes, box, mode = 'same')
|
|
return rate
|
|
|
|
def power_spectrum(stimulus):
|
|
'''
|
|
Computes a power spectrum based from a stimulus
|
|
|
|
Parameters
|
|
----------
|
|
stimulus : Stimulus object or rlxnix.base.repro module
|
|
The stimulus for which the data is needed.
|
|
|
|
Returns
|
|
-------
|
|
freq : np.array
|
|
All the frequencies of the power spectrum.
|
|
power : np.array
|
|
Power of the frequencies calculated.
|
|
|
|
'''
|
|
spikes, duration, dt = spike_times(stimulus)
|
|
# binarizes spikes
|
|
binary = binary_spikes(spikes, duration, dt)
|
|
# computes firing rates
|
|
rate = firing_rate(binary, dt = dt)
|
|
# creates power spectrum
|
|
freq, power = welch(rate, fs = 1/dt, nperseg = 2**16, noverlap = 2**15)
|
|
return freq, power
|
|
|
|
def remove_poor(files):
|
|
"""
|
|
Removes poor datasets from the set of files for analysis
|
|
|
|
Parameters
|
|
----------
|
|
files : list
|
|
list of files.
|
|
|
|
Returns
|
|
-------
|
|
good_files : list
|
|
list of files without the ones with the label poor.
|
|
|
|
"""
|
|
# create list for good files
|
|
good_files = []
|
|
# loop over files
|
|
for i in range(len(files)):
|
|
# print(files[i])
|
|
# load the file (takes some time)
|
|
data = rlx.Dataset(files[i])
|
|
# get the quality
|
|
quality = str.lower(data.metadata["Recording"]["Recording quality"][0][0])
|
|
# check the quality
|
|
if quality != "poor":
|
|
# if its good or fair add it to the good files
|
|
good_files.append(files[i])
|
|
return good_files
|
|
|
|
def sam_data(sam):
|
|
'''
|
|
Gets metadata for each SAM
|
|
|
|
Parameters
|
|
----------
|
|
sam : ReproRun object
|
|
The sam the metdata should be extracted from.
|
|
|
|
Returns
|
|
-------
|
|
sam_amp : float
|
|
amplitude in percent, relative to the fish amplitude.
|
|
sam_am : float
|
|
Amplitude modulation frequency.
|
|
sam_df : float
|
|
Difference from the stimulus to the current fish eodf.
|
|
sam_eodf : float
|
|
The current EODf.
|
|
sam_nyquist : float
|
|
The Nyquist frequency of the EODf.
|
|
sam_stim : float
|
|
The stimulus frequency.
|
|
|
|
'''
|
|
# create lists for the values we want
|
|
amplitudes = []
|
|
dfs = []
|
|
eodfs = []
|
|
stim_freqs = []
|
|
amp_mods = []
|
|
ny_freqs = []
|
|
|
|
# get the stimuli
|
|
stimuli = sam.stimuli
|
|
|
|
# loop over the stimuli
|
|
for stim in stimuli:
|
|
amplitude, df, eodf, stim_freq, amp_mod, ny_freq = extract_stim_data(stim)
|
|
amplitudes.append(amplitude)
|
|
dfs.append(df)
|
|
eodfs.append(eodf)
|
|
stim_freqs.append(stim_freq)
|
|
amp_mods.append(amp_mod)
|
|
ny_freqs.append(ny_freq)
|
|
|
|
# get the means
|
|
sam_amp = np.mean(amplitudes)
|
|
sam_am = np.mean(amp_mods)
|
|
sam_df = np.mean(dfs)
|
|
sam_eodf = np.mean(eodfs)
|
|
sam_nyquist = np.mean(ny_freqs)
|
|
sam_stim = np.mean(stim_freqs)
|
|
return sam_amp, sam_am,sam_df, sam_eodf, sam_nyquist, sam_stim
|
|
|
|
def spike_times(stim):
|
|
"""
|
|
Reads out the spike times and other necessary parameters
|
|
|
|
Parameters
|
|
----------
|
|
stim : Stimulus object or rlxnix.base.repro module
|
|
The stimulus from which the spike times should be calculated.
|
|
|
|
Returns
|
|
-------
|
|
spike_times : np.array
|
|
The spike times of the stimulus.
|
|
stim_dur : float
|
|
The duration of the stimulus.
|
|
dt : float
|
|
Time interval between two data points.
|
|
|
|
"""
|
|
# reads out the spike times
|
|
spikes, _ = stim.trace_data('Spikes-1')
|
|
# reads out the duration
|
|
stim_dur = stim.duration
|
|
# get the stimulus interval
|
|
ti = stim.trace_info("V-1")
|
|
dt = ti.sampling_interval
|
|
return spikes, stim_dur, dt # se changed spike_times to spikes so its not the same as name of function
|
|
|
|
'''TODO: AM-freq plot:
|
|
meaning of am peak in spectrum? why is it there how does it change with stim intensity?
|
|
make plot with AM 1/2 EODf over stim frequency (df+eodf), get amplitude of am peak and plot
|
|
amplitude over frequency of peak'''
|
|
|
|
|
|
def calculate_integral(freq, power, point, delta):
|
|
"""
|
|
Calculate the integral around a single specified point.
|
|
|
|
Parameters
|
|
----------
|
|
frequency : np.array
|
|
An array of frequencies corresponding to the power values.
|
|
power : np.array
|
|
An array of power spectral density values.
|
|
point : float
|
|
The harmonic frequency at which to calculate the integral.
|
|
delta : float
|
|
Half-width of the range for integration around the point.
|
|
|
|
Returns
|
|
-------
|
|
integral : float
|
|
The calculated integral around the point.
|
|
local_mean : float
|
|
The local mean value (adjacent integrals).
|
|
"""
|
|
indices = (freq >= point - delta) & (freq <= point + delta)
|
|
integral = np.trapz(power[indices], freq[indices])
|
|
|
|
left_indices = (freq >= point - 5 * delta) & (freq < point - delta)
|
|
right_indices = (freq > point + delta) & (freq <= point + 5 * delta)
|
|
|
|
l_integral = np.trapz(power[left_indices], freq[left_indices])
|
|
r_integral = np.trapz(power[right_indices], freq[right_indices])
|
|
|
|
local_mean = np.mean([l_integral, r_integral])
|
|
return integral, local_mean
|
|
|
|
|
|
|
|
def valid_integrals(integral, local_mean, threshold, point):
|
|
"""
|
|
Check if the integral exceeds the threshold compared to the local mean and
|
|
provide feedback on whether the given point is valid or not.
|
|
|
|
Parameters
|
|
----------
|
|
integral : float
|
|
The calculated integral around the point.
|
|
local_mean : float
|
|
The local mean value (adjacent integrals).
|
|
threshold : float
|
|
Threshold value to compare integrals with local mean.
|
|
point : float
|
|
The harmonic frequency point being evaluated.
|
|
|
|
Returns
|
|
-------
|
|
valid : bool
|
|
True if the integral exceeds the local mean by the threshold, otherwise False.
|
|
message : str
|
|
A message stating whether the point is valid or not.
|
|
"""
|
|
valid = integral > (local_mean * threshold)
|
|
if valid:
|
|
message = f"The point {point} is valid, as its integral exceeds the threshold."
|
|
else:
|
|
message = f"The point {point} is not valid, as its integral does not exceed the threshold."
|
|
return valid, message
|