pmt() function too slow - here are some ways to make it faster about numpy-financial HOT 3 OPEN

@garfieldthecat I'm sorry that no one has taken the time to look at this. I've started doing some maintenance work on numpy-financial and have also noticed the speed to be an issue for some functions. At a glance, Numba seems to be a good fit here.

I'm currently in the process of adding asv benchmarks, once that's done would you still be interested in contributing?

from numpy-financial.

Sure, do let me know. A couple of comments:

1. part of the reason why all the functions are very slow seems to be the fact that they are set up to accept both scalars and arrays as inputs. I have coded my own functions which accept only scalars and they're so much faster

2. Numba is not even at version 1. I am not sure everyone can/wants to install it. By all means, do considera a numba implementation, but maybe as an option? If you make your functions rely on numba only it might be tricky

from numpy-financial.

1. part of the reason why all the functions are very slow seems to be the fact that they are set up to accept both scalars and arrays as inputs. I have coded my own functions which accept only scalars and they're so much faster

The functions have been set up to act similarly to numpy ufuncs. I believe this can now be achieved using np.vectorize or np.fromfunc. Perhaps these are an avenue worth exploring?

There are more considerations, off the top of my head, this includes:

1. Do they handle masked arrays
2. Handling when (this could be expensive if vectorized)
3. Handling Decimal datatype

2. Numba is not even at version 1. I am not sure everyone can/wants to install it. By all means, do considera a numba implementation, but maybe as an option? If you make your functions rely on numba only it might be tricky

Good point. Numba was suggested to me by another NumPy developer and it's been sitting in the back of my mind while I've been working. Either way I'd prefer to have some full benchmarks before anything changes in the name of performance.

from numpy-financial.