anaties

Anaties (contraction of 'analysis utilities'). A place for common operations like signal smoothing that are useful across all my data analysis projects.

Installation and usage

Install with pip:

pip install anaties

When a new release is made, upgrade with:

pip install anaties --upgrade

Usage is simple. In your code:

import anaties as ana
ana.function_name()

You can test it out with:

import anaties as ana
print(ana.datetime_string())

plt.plot([0, 1], [0,1], color='k', linewidth=0.6)
ana.rect_highlight([0.25, 0.5])

All other functions are listed below.

Brief summary of all utilities

    signals.py (for 1d data arrays, or arrays of such arrays)
        - smooth: smooth a signal with a window (gaussian, etc)
        - smooth_rows: smooth each row of a 2d array using smooth()
        - power_spec: get the power spectral density or power spectrum
        - spectrogram: calculate/plot spectrogram of a signal
        - notch_filter: notch filter to attenuate specific frequency (e.g. 60hz)
        - bandpass_filter: allow through frequencies within low- and high-cutoff

    plots.py (basic plotting)
        - error_shade: plot line with shaded error region
        - freqhist: calculate/plot a relative frequency histogram
        - paired_bar: bar plot for paired data
        - plot_with_events: plot with vertical lines to indicate events
        - rect_highlight: overlay rectangular highlight to figure
        - vlines: add vertical lines to figure

    stats (basic statistical things)
        - collective_correlation: collective correlation coefficient
        - med_semed: median and std error of median of an array
        - mean_sem: mean and std error of the mean of an array
        - mean_std: mean and standard deviation of an array
        - se_mean: std err of mean of array
        - se_median: std error of median of array
        - cramers_v: cramers v for effect size for chi-square test

    helpers.py (generic utility functions for use everywhere)
        - datetime_string : return date_time string to use for naming files etc
        - file_exists: check to see if file exists
        - get_bins: get bin edges and centers, given limits and bin width
        - get_offdiag_vals: get lower off-diagonal values of a symmetric matrix
        - ind_limits: return indices that contain array data within range
        - is_symmetric: check if 2d array is symmetric
        - rand_rgb: returns random array of rgb values

Acknowledgments

• Songbird wav is open source from: https://freesound.org/people/Sonic-ranger/sounds/243677/
• Developed with the support from NIH Bioinformatics and the Neurobehavioral Core at NIEHS.

To do: More important

• finish adding tests.
• plots.rect_highlight should just use axvspan/axhspan!
• use median instead of mean in spectrogram
• add proper documentation and tests to stats module.
• integrate vlines into pypi and version up (maybe good test for ci)
• add ax return for all plot functions, when possible.
• finish plots.twinx and make sure it works
• add test for plots.error_shade.
• Add return object for plots.rect_highlight()
• consider adding directory_exists to helpers
• paired_bar and mean_sem/std need to handle one point better (throws warning)
• Add a proper suptitle fix in aplots it is a pita to add manually/remember: f.suptitle(..., fontsize=16) f.tight_layout() f.subplots_adjust(top=0.9)
• For freqhist should I guarantee it sums to 1 even when bin widths don't match data limits? Probably not. Something to think about though.
• In smoother, consider switching from filtfilt() to sosfiltfilt() for reasons laid out here: https://dsp.stackexchange.com/a/17255/51564
• Convert notch filter to sos?
• For spectral density estimation consider adding multitaper option. Good discussions: https://github.com/cokelaer/spectrum https://pyspectrum.readthedocs.io/en/latest/ https://mark-kramer.github.io/Case-Studies-Python/04.html
• add ability to control event colors in spectrogram.
• ind_limits: add checks for data, data_limits, clarify description and docs
• Add numerical tests with random seed set not just graphical eyeball tests.

To do: longer term

• Add audio playback of signals (see notes in audio_playback_workspace), incorporate this into some tests of filtering, etc.. simpleaudio package is too simple I think.
• autodocs (sphinx?)
• CI/CD with github actions
• consider adding wavelets.
• Add 3d array support for stat functions like mn_sem

Useful sources

Smoothing

• https://scipy-cookbook.readthedocs.io/items/FiltFilt.html
• https://docs.scipy.org/doc/scipy/reference/generated/scipy.signal.filtfilt.html

What about wavelets?

I may add wavelets at some point, but it isn't plug-and-play enough for this repo. If you want to get started with wavelets in Python, I recommend http://ataspinar.com/2018/12/21/a-guide-for-using-the-wavelet-transform-in-machine-learning/

Tolerance values

For a discussion of the difference between relative and absolute tolerance values when testing floats for equality (for instance as used in helpers.is_symmetric()) see: https://stackoverflow.com/questions/65909842/what-is-rtol-for-in-numpys-allclose-function

Suggestions?

If there is something you'd like to see, please open an issue.