Feature/graphwave chebyshev #920
Conversation
- refactored embedding calculation - added `scales="auto"` option
# Conflicts: # stellargraph/mapper/__init__.py
…d added GraphWave info into the readme
Co-Authored-By: Huon Wilson <Huon.Wilson@data61.csiro.au>
…o feature/graphwave-chebyshev # Conflicts: # tests/mapper/test_graphwave_generator.py
![codeclimate[bot]](https://avatars.githubusercontent.com/in/9403?s=60&v=4){kind=small}
| (num_scales, num_nodes) tensor of the wavelets for each scale for the specified node. | ||
| """ | ||
|
|
||
| def y(vector): |
There was a problem hiding this comment.
Rename function "y" to match the regular expression ^[a-z_][a-z0-9_]{2,}$.
| """ | ||
| Args: | ||
| G (StellarGraph): the StellarGraph object. | ||
| scales (iterable of floats): the wavelet scales to use. Smaller values embed smaller scale structural | ||
| features, and larger values embed larger structural features. | ||
| deg: the degree of the Chebyshev polynomial to use. Higher degrees yield more accurate results but at a | ||
| higher computational cost. The default value of `20` is accurate enough for most applications. |
There was a problem hiding this comment.
Maybe we could note it, so that users (and us) can remember where the assertion comes from:
| higher computational cost. The default value of `20` is accurate enough for most applications. | |
| higher computational cost. According to [1], the default value of `20` is accurate enough for most applications. |
With a citation somewhere in this doc string:
[1] D. I. Shuman, P. Vandergheynst, and P. Frossard, “Chebyshev Polynomial Approximation for Distributed Signal Processing,” https://arxiv.org/abs/1105.1891
| cheby_polys.append(cheby_poly) | ||
|
|
||
| cheby_polys = K.squeeze(tf.stack(cheby_polys, axis=0), axis=-1) | ||
| return tf.matmul(coeffs, cheby_polys) |
There was a problem hiding this comment.
It would be good to note, because it's different to the paper we're using for reference.
# Conflicts: # stellargraph/mapper/graphwave_generator.py
There was a problem hiding this comment.
Only a few very tiny things, and then I think it's good.
Also, it's a bit buried now, but I replied to #920 (comment) .
stellargraph/core/validation.py
Outdated
| min_val: the minimum value that `x` can attain | ||
| min_val: the maximum value that `x` can attain | ||
| """ | ||
| if min_val == -np.inf: |
There was a problem hiding this comment.
It would be very slightly better for this to be inside the if x < min_val or ...: branch, so that we don't waste time doing these checks or string formatting if x is valid.
Co-Authored-By: Huon Wilson <Huon.Wilson@data61.csiro.au>
Co-Authored-By: Huon Wilson <Huon.Wilson@data61.csiro.au>
# Conflicts: # CHANGELOG.md
This PR adds unit tests and Chebyshev polynomials for GraphWave, and removes the experimental tag from GraphWave.
Chebyshev polynomials are used approximate spectral graph filtering for GraphWave instead of calculating all the eigenvalues explicitly. This decreases the computational complexity from
O(N^3)toO(kEn)where:Nis the number of nodeskis the order of the Chebyshev polynomialEis the number of edges in the graphnis the number of nodes to calculate embeddings for (length ofnode_idspassed togenerator.flowin this caseReviewer checklist:
test_chebyshevby comparing the approximation to the exact value)