Add fourier #49
Add fourier #49
Conversation
| # ### The Fourier Basis | ||
| # Another type of basis available is the Fourier Basis. Fourier basis are ideal to capture periodic and | ||
| # quasi-periodic patterns. Such oscillatory, rhythmic behavior is a common signature of many neural signals. | ||
| # Additionally, the Fourier basis has the advantage of being orthogonal, which simplifies the estimation and |
There was a problem hiding this comment.
Should probably explain orthogonal here.
There was a problem hiding this comment.
still think this. at least a foot note or link
Co-authored-by: William F. Broderick <billbrod@gmail.com>
src/nemos/basis.py
Outdated
| """ | ||
| if self.n_basis_funcs < 1: | ||
| raise ValueError( | ||
| f"Object class {self.__class__.__name__} requires >= 1 basis elements. " | ||
| f"Object class {self.__class__.__name__} requires >= 0 basis elements. " |
There was a problem hiding this comment.
I see the error message now matches the check, but shouldn't this actually at least 1 basis element? With max_freq=0 (just the DC term), you get a n_basis_funcs=1. And it doesn't make sense to have max_freq<0
| # ### The Fourier Basis | ||
| # Another type of basis available is the Fourier Basis. Fourier basis are ideal to capture periodic and | ||
| # quasi-periodic patterns. Such oscillatory, rhythmic behavior is a common signature of many neural signals. | ||
| # Additionally, the Fourier basis has the advantage of being orthogonal, which simplifies the estimation and |
There was a problem hiding this comment.
still think this. at least a foot note or link
| # Note that we are inverting the time axis of the basis because we are aiming | ||
| # for a cross-correlation, while np.convolve compute a convolution which would flip the time axis. |
| # compute the phase and amplitude from the convolution | ||
| xcorr_phase = np.arctan2(np.hstack([[0], xcorr[n_freqs:]]), xcorr[:n_freqs]) | ||
| xcorr_aplitude = np.sqrt(xcorr[:n_freqs] ** 2 + np.hstack([[0], xcorr[n_freqs:]]) ** 2) |
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## main #49 +/- ##
==========================================
+ Coverage 95.02% 95.34% +0.32%
==========================================
Files 10 10
Lines 864 881 +17
==========================================
+ Hits 821 840 +19
+ Misses 43 41 -2 ☔ View full report in Codecov by Sentry. |
| >>> import nemos as nmo | ||
| >>> import numpy as np | ||
| >>> n_samples, max_freq = 1000, 10 | ||
| >>> basis = nmo.basis.FourierBasis(max_freq) | ||
| >>> eval_basis = basis.evaluate(np.linspace(0, 1, n_samples)) | ||
| >>> sinusoid = np.cos(3 * np.arange(0, 1000) * np.pi * 2 / 1000.) | ||
| >>> conv = [np.convolve(eval_basis[::-1, k], sinusoid, mode='valid')[0] for k in range(2*max_freq+1)] | ||
| >>> fft = np.fft.fft(sinusoid) | ||
| >>> print('FFT power: ', np.round(np.real(fft[:max_freq]), 4)) | ||
| >>> print('Convolution: ', np.round(conv[:max_freq], 4)) |
There was a problem hiding this comment.
Is there a way in mkdocs to set this as a python codeblock? Can you remove the >>>? Because right now in the docs the >>> shows up and when you copy it copies with the >>> so a user can't just copy and paste and run from the docs 😄
for example at the bottom here the arrows render and get copied: https://nemos.readthedocs.io/en/latest/reference/nemos/utils/#nemos.utils.pytree_map_and_reduce
| (sample_pts,) = self._check_evaluate_input(sample_pts) | ||
| # assumes equi-spaced samples. | ||
| if sample_pts.shape[0] / np.max(self._frequencies) < 2: | ||
| raise ValueError("Not enough samples, aliasing likely to occur!") |
There was a problem hiding this comment.
Maybe report sample_pts.shape[0] and max(self._frequencies) to the user?
Co-authored-by: William F. Broderick <billbrod@gmail.com>
Co-authored-by: William F. Broderick <billbrod@gmail.com>
FourierBasisto the basis module