How to use vector in a numba function? #206
Answered
by
jpivarski
ZhenxuanZhang-Jensen
asked this question in
Q&A
-
|
just a very simple code to add two jets to form a W boson, if I just use a common function, it can run successfully but it can't be done in Numba: LV_jets = ak.zip({"pt": [1,2,3], "phi": [4,5,6], "eta": [7,8,9], "M": [10,11,12]}, with_name="Momentum4D")
@nb.njit
def SortWithWMass(LV_jets):
print(LV_jets[0] + LV_jets[1])
SortWithWMass(LV_jets)And the bug show this: |
Beta Was this translation helpful? Give feedback.
Answered by
jpivarski
Jun 16, 2022
Replies: 1 comment 4 replies
-
|
Hmmm... I just tried this and it worked: >>> import awkward as ak
>>> import numba as nb
>>> LV_jets = ak.zip({"pt": [1,2,3], "phi": [4,5,6], "eta": [7,8,9], "M": [10,11,12]}, with_name="Momentum4D")
>>>
>>> @nb.njit
... def SortWithWMass(LV_jets):
... print(LV_jets[0] + LV_jets[1])
...
>>> SortWithWMass(LV_jets)
vector.obj(pt=2.6760435765272135, phi=-1.603058217614354, eta=7.877655297214794, mass=28.12262855227671)The error message above is saying that it doesn't recognize the objects that are being combined with and then it desperately tries to figure out why, producing a lot of output in the process. Could you try explicitly registering Awkward Array and Vector with Numba? >>> ak.numba.register()
>>> vector.register_numba()These functions are supposed to be automatically called when |
Beta Was this translation helpful? Give feedback.
4 replies
-
|
Hi Jim, I rerun the code like this: import awkward as ak
import numba as nb
import vector
ak.numba.register()
vector.register_numba()
LV_jets = ak.zip({"pt": [1,2,3], "phi": [4,5,6], "eta": [7,8,9], "M": [10,11,12]}, with_name="Momentum4D")
@nb.njit
def SortWithWMass(LV_jets):
print(LV_jets[0] + LV_jets[1])
SortWithWMass(LV_jets)the version I use for awkward, vector, Numba are 1.8.0, 0.8.5, and 0.51.2. And it shows the bug: |
Beta Was this translation helpful? Give feedback.
-
|
Okay, I should have looked more deeply. At the most indented level (the point at which Numba finally gave up), it says
There are a lot of issues on the Numba GitHub about this:
The answers from the Numba developers are "we're not ready for Python 3.9 yet." But those are historical answers; I know that Numba has been updated for Python 3.10 already. Your Numba version, 0.51.2, was released on September 3, 2020. Your Python version, 3.9 was released on October 5, 2020. So your Numba version predates your Python version, and Python is producing opcodes (IS_OP) that Numba does not recognize. That all was fixed years ago, and so updating your Numba version should fix this error. (Is it necessary to run on Python code distributed by LCG? Could you set yourself up with a conda-forge installation and use modern versions of all these packages, all installed from a single distribution?) |
Beta Was this translation helpful? Give feedback.
-
|
ahh, okay, I upgrade my =Numba version and it works... I have some reasons to use LCG: 1. I want to use swan so that I can share my project with my CERN teammates easily. 2. I try to use conda-forge installation to set all my environments and I meet some issues in different ways: (1) When I set in my own computer, it turns out ROOT/Pyroot can't run on any M1 chip mac computer(I tried this half year ago, maybe it's fixed now). (2) When I set in the CERN server, it was very slow to run, and also it never recognize the right kernel for me on jupyter notebook. So after that, I find SWAN is the most comfortable environment for me to use, at least for now... |
Beta Was this translation helpful? Give feedback.
-
|
Okay, makes sense. The only thing that seems to be hard with SWAN, though, is the installation of packages. In a conda-only environment, |
Beta Was this translation helpful? Give feedback.
Answer selected by
ZhenxuanZhang-Jensen
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Hmmm... I just tried this and it worked:
The error message above is saying that it doesn't recognize the objects that are being combined with
+: