Faster hermite #280
Faster hermite #280
Conversation
|
@ziofil --- Could you update the CHANGELOG? Also is this giving a speed increase? If so it would be nice to report it. |
Do you have a preferred speed test to compare it against? |
|
You could run the basic calculation that we use to test the correctness of the numba hermite vs the C++ hermites? |
Sure, where do I find it? |
|
In the |
|
cutoff = 5 and 1,2,3,4,5 indices
|
|
sped it up a bit |
| def _grad_hermite_multidimensional_numba(R, y, array, dG_dR, dG_dy, cutoffs): | ||
| indices = np.ndindex(cutoffs) | ||
| next(indices) # skip the first index (0,...,0) |
There was a problem hiding this comment.
can you add a docstring here?
| def _hermite_multidimensional_numba(R, array, y, cutoffs): | ||
| indices = np.ndindex(cutoffs) |
There was a problem hiding this comment.
same here, a docstring would be nice
Codecov Report
@@ Coverage Diff @@
## master #280 +/- ##
=========================================
Coverage 100.00% 100.00%
=========================================
Files 21 21
Lines 1441 1428 -13
=========================================
- Hits 1441 1428 -13
Continue to review full report at Codecov.
|
🤯 How is Numba able to do this? |
probably the C++ code is not well written 😅 |
fully numbified filling of termite multitimensional arrays and gradients. It turns out
np.ndindexis supported in numba.