Improving your IRASA model fits#

When fitting IRASA we sometimes face the situation that residual periodic activity remains in our aperiodic spectrum. In this notebook I will illustrate two possible ways that could lead to this situation and how to avoid it. First we start again by simulating a fairly simple signal consisting of a 10Hz oscillation and an aperiodic exponent of 1.

[1]:

from neurodsp.sim import sim_combined
from neurodsp.utils import create_times
import numpy as np
import scipy.signal as dsp
import matplotlib.pyplot as plt
import pandas as pd
from pyrasa.irasa import irasa

[2]:

fs = 500
n_seconds = 60

sim_components = {'sim_powerlaw': {'exponent' : -1},
                  'sim_oscillation': {'freq' : 10}}


sig = sim_combined(n_seconds=n_seconds, fs=fs, components=sim_components)
times = create_times(n_seconds=n_seconds, fs=fs)

Case #1: Increase your frequency resolution#

When the frequency resolution for your psd is “low” up/-downsampling is not enough to reliably get rid of peaks in the signal. As can be seen below where a duration of 1s was not enough to obtain a “peak-free” spectrum.

[12]:

#low segment length
duration=1
overlap=0.5
irasa_out = irasa(sig,
                    fs=fs,
                    band=(1, 100),
                    psd_kwargs={'nperseg': int(duration*fs),
                                'noverlap': int(duration*fs*overlap)
                            },
                    hset_info=(1, 2, 0.05))

plt.loglog(irasa_out.freqs, irasa_out.aperiodic.T)

irasa_out.fit_aperiodic_model().aperiodic_params

[12]:
Offset Exponent fit_type ch_name
0 -1.168134 1.037697 fixed 0
../_images/examples_improving_irasa_models_4_1.png

Increasing the length of each segment (i.e. the “duration”) should help you out.

[13]:

#Increased segment length
duration=6
overlap=0.5
irasa_out = irasa(sig,
                    fs=fs,
                    band=(1, 100),
                    psd_kwargs={'nperseg': int(duration*fs),
                                'noverlap': int(duration*fs*overlap),
                            },
                    hset_info=(1, 2, 0.05))

plt.loglog(irasa_out.freqs, irasa_out.aperiodic.T)

irasa_out.fit_aperiodic_model().aperiodic_params

[13]:
Offset Exponent fit_type ch_name
0 -1.25566 0.99467 fixed 0
../_images/examples_improving_irasa_models_6_1.png

Case #2: Increase your hmax#

Sometimes we either can’t or don’t want to increase our frequency resolution. In this case we can increase our maximal up/-downsampling factors to get rid of peaks in the signal. However, be aware that this can - depending on our filter settings and sampling rate - limit the frequency range that we can sensibly evaluate (see irasa_pitfalls.ipynb).

[11]:

#low segment length, but high hset
duration=1
overlap=0.5
irasa_out = irasa(sig,
                    fs=fs,
                    band=(4, 40),
                    psd_kwargs={'nperseg': int(duration*fs),
                                'noverlap': int(duration*fs*overlap)
                            },
                    hset_info=(1.1, 6, 0.05))

plt.loglog(irasa_out.freqs, irasa_out.aperiodic.T)

irasa_out.fit_aperiodic_model().aperiodic_params

[11]:
Offset Exponent fit_type ch_name
0 -1.252254 0.991107 fixed 0
../_images/examples_improving_irasa_models_8_1.png 
[ ]:
On this page