jakevdp
Follow
jakevdp
Follow
Achievements
Achievements
Highlights
- 82 discussions answered
Pinned
1,399 contributions in the last year
Less
More
Contribution activity
October 2021
Created 47 commits in 2 repositories
Created a pull request in google/jax that received 5 comments
DOC: add deprecation message to index_update and friends
Fixes #8064 Preview the new docs here: https://jax--8065.org.readthedocs.build/en/8065/_autosummary/jax.ops.index_update.html#jax.ops.index_update
+57
−12
•
5
comments
Opened 43 other pull requests in 2 repositories
google/jax
3
open
36
merged
2
closed
- jnp.histogramdd: more succinct density computation
- dynamic_slice: ensure start_indices dtypes match
- odeint: validate *args
- [sparse] improve error for BCOO.fromdense if nse is not specified
- Add conv_dimension_numbers and ConvGeneralDilatedDimensionNumbers to docs
- special.lpmn: use more canonical testing approach
- stats.multivariate_normal: support broadcasted inputs
- [sparse] preserve dtype in bcoo_todense
- [sparse] respect weak types in sparsify transform
- Fix inaccurate type annotations
- jnp.bincount: fix corner cases & improve tests
- [sparse]: change bcoo pad values to use OOB indices
- jnp.unique: improve efficiency & consolidate implementation
- jnp.unique: allow fill_value to be a slice
- jnp.unique: don't apply fill_value to indices
- jnp.nonzero: allow fill_value to be a tuple
- jnp.piecewise: avoid unnecessary recompilation
- jnp.unique: support size argument with axis
- WIP: change dtype default to 32-bit
- jax.numpy: explicitly use dtypes.scalar_type when appropriate
- [sparse] fix padding bug in BCOO._dedupe()
- jnp.setdiff1d: add optional size and fill_value arguments
- jnp.union1d: add optional fill_value argument
- BUG: fix gradients for nanvar & nanstd
- jnp.array: handle raw device buffers
- Some pull requests not shown.
altair-viz/altair
2
merged
Reviewed 31 pull requests in 2 repositories
google/jax
30 pull requests
- add normalizations for FFT functions
- Fix broken link to experimental/README.md
- [sparse] improve error for BCOO.fromdense if nse is not specified
- Improve real type conversion in a couple more places.
- Remove unused backward compatibility code in cusolver.py.
-
Change default kind for jnp.argsort to
stable. Warn if anything oth… -
document
axis_namein thevmapdocstring - jnp.take/jnp.take_along_axis: require array inputs
- Readd jax.test_util.check_jvp and check_vjp to the public JAX API.
- jnp.piecewise: avoid unnecessary recompilation
- Fix CHANGELOG.md for #8161
-
Cleanup
random.permutation - Remove unused variable from notebooks
- Remove the _num_buffers attribute from core.AbstractValue.
- jnp.unique: support size argument with axis
-
Allow
random.choiceandrandom.permutationon multidimensional arrays - fix #8152
- add common virtualenv directories to .gitignore
- jnp.setdiff1d: add optional size and fill_value arguments
- jnp.array: handle raw device buffers
- Update doc for eigh.
- Document extra arguments to jnp.ndarray.at[]
- jnp.array: replace host round-trip with on-device copy
- Add DeprecationWarnings to jax.ops.index_... operators.
- Fix incorrect EllipsisType reference for Python 3.10
- Some pull request reviews not shown.
scikit-learn/scikit-learn
1 pull request
Created an issue in google/jax that received 8 comments
Enhancement: support normalization options in FFT
numpy 1.20 and later supports several alternate normalizations for FFT functions, including "backward", "ortho", and "forward". We should support t…
8
comments
Opened 7 other issues in 2 repositories
google/jax
3
open
3
closed
scipy/scipy
1
open
Answered 14 discussions in 1 repository
google/jax
google/jax
- Installing jaxlib
- vmap with non vmappable parameter
- Maximum number of iteration in lax.while_loop
- A mistake(?) in Std implementation
- Help with learning
- Help with learning
- Is there a jax equivalent to numpy.mod.outer?
- How to use jax.random package for sampling general normal distribution?
- How does JAX PRNG improve on Mersenne Twister beyond splitting and explicit key?
- Manually inspecting jitted code
- jax.numpy.nanstd bad behaviour with jax.grad?
- Zero class
- transpose over batch of 3d arrays using vmap
- jnp.where and NaN in gradients