I am a machine learning research, using the python/scipy/theano stack. One problem performance wise is that I can get my models reasonably fast with Theano--but as soon as I need a python for loop (e.g. by using Theano.scan) performance dies. As soon as you have a couple of nested (e.g. due to a Jacobian matrix instead of vector) performance is virtually gone again.
I'd love to use numpypy, and the way I see it not much is missing.
This is good news, although confusing Laplace transform and Laplace equation makes me think that before using their numeric code I should check it thoroughly.
Fortunately for you, we're only dealing with things like addition, multiplication, maybe sinus and even so we rely on the processor to do the job :-) The rest is dealt by numeric guys from the original numpy and we're simply reusing those parts. It's important for us to understand how processors work and how to make sure we do exactly the same computations, but not much above that.
Besides, believe it or not, having non-english maths background makes you seem incredibly dumb (which might or might not be the case).
EDIT: I should maybe stress this point more. It's very important for us to get exactly the same results, or more numerically stable as original numpy, not just the same algorithms, so we won't be experimenting on that field.
I am a machine learning research, using the python/scipy/theano stack. One problem performance wise is that I can get my models reasonably fast with Theano--but as soon as I need a python for loop (e.g. by using Theano.scan) performance dies. As soon as you have a couple of nested (e.g. due to a Jacobian matrix instead of vector) performance is virtually gone again.
I'd love to use numpypy, and the way I see it not much is missing.
Keep up the good work! I wish I cold fund you! :/