Fixes #7 via sketch module. #19
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…odule that incorporates additional sketching aglorithms
Work in progress. ``sketch`` module is implemented, but old compatibility funcitons in ``sketch.range_finders`` still need to be phased out.
Still need to better incorporate within random methods, perhaps giving transform options. Also still need to standardize parameter names. Additionally, single_pass methods have been depricated (masked with a single subspace iteration) for now because the previous implementation was not in fact 'single_pass'.
For the most part package now conforms with PEP parameter name guidelines. Still to do: - `pca` : factor pca/spca/robspca - `nmf` : factor and PEP parameter name eval - `sketch` : ensure row wise sketches are indeed row wise and write numerical accuracy checking tests Additionally, in general there can be a lot of improvement as far as code reproduction reduction in the testing modules.
The calling of `perform_subspace_iterations` now resides outside of the transform functions so that it can be used independent of which transform is used. Additionally, default n_subspace has been set to 2.
erichson
reviewed
May 25, 2018
| # most definitely to sparsely sample the nnz elements of Omega, but | ||
| # is using random_state in data_rvs redundant? | ||
| values = (-sqrt(1. / density), sqrt(1. / density)) | ||
| data_rvs = partial(random_state.choice, values) |
There was a problem hiding this comment.
Random_state is not redundant here.
|
|
||
| Parameters | ||
| ---------- | ||
| A : array_like, shape `(n, n)`. | ||
| Positive-definite matrix (PSD) input matrix. |
There was a problem hiding this comment.
We should mention that a PSD matrix is required, maybe just us a note.
There was a problem hiding this comment.
Yes we should. We should also check by default that the matrix is indeed PSD, but have an option not to (similar to check_finite in scipy.linalg.qr)
| return safe_sparse_dot(A, Omega) | ||
|
|
||
|
|
||
| def fast_johnson_lindenstrauss(A, l, axis=1, random_state=None): |
There was a problem hiding this comment.
Awesome!! Excited to see the speedups!
|
It looks like the Travis build failed because of a failed install of Python 3.5 (some weird travis bug), but after rerunning, the build passes! |
|
@erichson Does this look good to you? |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There are 2 main updates with this PR:
sketchmodule as a subpackage containing sketching routines (_sketch) as well as transform algorithms (transforms)q -> n_subspace)Also, I have removed the
single_passalgorithms, because their use (only accessingAa single time) was not yet implemented: we computedA.dot(Omega)andPsi.dot(A), instead of doing both in a single pass overA. We can look back into this later, though it would require writing a C/Cython or Fortran subroutine.The next step I think is implementing
sparseoptions for sparse sketching inside the factorization algorithms to solve #11, but I would like to make a separate PR for takling this.