nelpy/ghost

Spectral analysis tools for neuroscience data

ghost (Grand Hub Of Spectral Tools) is the friendly phantom that helps you do signal processing on neuroscience data, especially spectral analysis ;)

It can be used as a standalone package for numpy arrays, or as a companion plugin module to nelpy. Currently it supports rudimentary wavelet analysis. Planned features include multitaper Fourier methods, phase-amplitude & phase-phase coupling, current source density analysis, and more.

ghost tries to be as lightweight yet fast as possible. Suggestions for increasing efficiency and performance are always welcome!

Installation

To install this package, please run the following commands

$ git clone https://github.com/nelpy/ghost.git
$ cd ghost
$ pip install

If you are a developer and would like to modify the code, replace the last command above with

$ pip install -e .

Example

Suppose you have a numpy array named X which was sampled at 1 kHz, on which you want to run a continuous wavelet transform:

from ghost.wave import ContinuousWaveletTransform

cwt = ContinuousWaveletTransform()
cwt.transform(X, fs=1000)

If you have a nelpy AnalogSignalArray named asa, you can simply do:

from ghost.wave import ContinuousWaveletTransform

cwt = ContinuousWaveletTransform()
cwt.transform(asa)

In either case, you can obtain the spectrogram by calling plot:

cwt.plot(logscale=False, 
         standardize=True, 
         cmap=plt.cm.Spectral_r,
         levels=300, 
         vmin=0, 
         vmax=10)

And that's it! Short and simple.

ghost's Issues

Handle timestamp info for wavelet spectrogram plot

Currently the timestamps are generated from the data but there should be a way to incorporate optional passed-in timestamps for plotting a wavelet spectrogram.

Can't import ContinuousWaveletTransform with newest scipy

On google collaboratory, if I install ghost from git, and then run

from ghost.wave import ContinuousWaveletTransform

I get:

---------------------------------------------------------------------------

ImportError                               Traceback (most recent call last)

<ipython-input-8-e9048c87e447> in <module>()
----> 1 from ghost.wave import ContinuousWaveletTransform

2 frames

/content/ghost/ghost/wave/morlet.py in <module>()
      3 import numpy as np
      4 from .wavelet import Wavelet
----> 5 from scipy.misc import logsumexp
      6 import copy
      7 

ImportError: cannot import name 'logsumexp'

scipy fftpack missing

Once I fix the logsumexp issue, I get a new error:

from ghost.wave import ContinuousWaveletTransform

Now I get:

---------------------------------------------------------------------------

ModuleNotFoundError                       Traceback (most recent call last)

<ipython-input-5-e9048c87e447> in <module>()
----> 1 from ghost.wave import ContinuousWaveletTransform

5 frames

/content/ghost/ghost/sigtools/fourier.py in <module>()
      3 import logging
      4 import numpy as np
----> 5 import scipy.fftpack._fftpack as sff
      6 from . import convolution
      7 

ModuleNotFoundError: No module named 'scipy.fftpack._fftpack'

scipy: if _is_safe_size(n):

Hi,
This is my code with Accz and Time are array of float:

cwt = ContinuousWaveletTransform(wavelet= morse.Morse())
CPEF=cwt.transform(data =Accz, fs =1000, timestamps =Time, freq_limits =[0,300])

And the answer:

File ".......\lib\site-packages\scipy\fftpack\basic.py", line 87, in fft
if _is_safe_size(n):

TypeError: c2c() missing 1 required keyword-only argument: 'plan'

Adaptive sampling of wavelets

At the lower frequencies, we need more samples to get an accurate representation of the wavelet, but at higher frequencies, we need fewer. I need to figure out a better way of sampling the wavelets rather than the default of 16000 samples or so.

Too many points in wavelet spectrogram plot

If there are too many points in the wavelet spectrogram plot, some parts of the image will not render properly. I need to figure out a way to downsample correctly without messing up the temporal resolution.