SciPy LowLevelCallable from Lambdify #201
Conversation
|
A simple example, In [1]: import symengine as se
...: import numpy as np
...: from scipy import integrate
...: args = t, x = se.symbols("t, x")
...:
...: lmb = se.lambdify(args, [se.exp(-x*t)/t**5], as_scipy=True)
...: lmb2 = se.lambdify(args, [se.exp(-x*t)/t**5], as_scipy=False)
...:
In [2]: %timeit integrate.nquad(lmb, [[1, np.inf],[0, np.inf]])
1.39 ms ± 1.87 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)
In [3]: %timeit integrate.nquad(lmb2, [[1, np.inf],[0, np.inf]])
51.5 ms ± 425 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)cc @bjodah |
symengine/lib/symengine_wrapper.pyx
Outdated
| @@ -4579,6 +4604,16 @@ cdef class LambdaDouble(_Lambdify): | |||
| cpdef unsafe_complex(self, double complex[::1] inp, double complex[::1] out, int inp_offset=0, int out_offset=0): | |||
| self.lambda_double_complex[0].call(&out[out_offset], &inp[inp_offset]) | |||
|
|
|||
| cpdef as_scipy_low_level_callable(self): | |||
| from scipy import LowLevelCallable | |||
There was a problem hiding this comment.
is the import of LowLevelCallable needed here?
symengine/lib/symengine_wrapper.pyx
Outdated
| @@ -4592,8 +4627,18 @@ IF HAVE_SYMENGINE_LLVM: | |||
| cpdef unsafe_real(self, double[::1] inp, double[::1] out, int inp_offset=0, int out_offset=0): | |||
| self.lambda_double[0].call(&out[out_offset], &inp[inp_offset]) | |||
|
|
|||
| cpdef as_scipy_low_level_callable(self): | |||
| from scipy import LowLevelCallable | |||
|
This is really exciting and impressive (as always). I haven't used scipy's LowLevelCallable myself yet (I have been meaning to look into it). Right now I'm quite busy, but I can take a closer look at this PR this weekend. |
|
|
||
| def create_low_level_callable(lambdify, *args): | ||
| from scipy import LowLevelCallable | ||
| class LambdifyLowLevelCallable(LowLevelCallable): |
There was a problem hiding this comment.
Does this class need to be closure? Or could it be defined in a try/except block around from scipy import LowLevelCallable?
There was a problem hiding this comment.
I didn't want to add the initialization time of scipy to symengine. That's why this is done inside a function.
|
Looks good to me. Too bad the LowLevelCallable isn't used much in scipy at the moment (that might change though). |
|
Thanks for the review |
|
Very impressive. Thanks @isuruf ! |
|
Hi above all, thank you for this awesome tool, it will almost definitly do what i need. i have Best wishes Mustafa |
|
@BlackTeaAndCoffee, you'll have to use the latest version of symengine.py and symengine for this |
cc @BlackTeaAndCoffee
This creates a
LowLevelCallablethat can be used by scipy integrate functions without the overhead of python.