Numba won't accept numpy array datatype

Question

I have following dataframe:

                      time         t            sp
598258 2017-01-02 00:00:00 -2.634766  89892.492188
598259 2017-01-02 01:00:00 -2.753906  89921.398438
598260 2017-01-02 02:00:00 -2.730469  89896.914062
598261 2017-01-02 03:00:00 -2.765625  89874.468750
598262 2017-01-02 04:00:00 -2.855469  89864.609375
598263 2017-01-02 05:00:00 -3.005859  89846.929688
598264 2017-01-02 06:00:00 -3.011719  89877.875000
598265 2017-01-02 07:00:00 -2.900391  89873.109375
598266 2017-01-02 08:00:00 -2.416016  89891.812500
598267 2017-01-02 09:00:00 -2.126953  89882.289062

and following code:

import pandas as pd
import numpy as np
from numba import jit
@jit(nopython=True)
def en(t,p):
    dd = np.empty((t.size),np.float16)
    for i in range(len(t)):
        dd[i]=p[i]/287.05/(t[i]+273.15)
    return dd

ex=en(dfx['t'].values,dfx['sp'].values)

I get error:

TypingError: non-precise type pyobject
[1] During: typing of argument at dd = np.empty((t.size),np.float16)

I know I must define precise type for numba to accept it, i think I did just that with np.float16 but error is present. Any help to resolve this is appreciated

@Ethan Tried that, still same error.

user2727167
– user2727167

2019-12-22 19:36:51 +00:00
Commented Dec 22, 2019 at 19:36 — user2727167
– user2727167, Commented Dec 22, 2019 at 19:36

oppressionslayer · Accepted Answer · 2019-12-22 19:49:35Z

2

You have to change to float 64 or 32 depending on the precision you want:

from numba import jit
@jit(nopython=True)
def en(t,p):
    dd = np.empty((t.size),np.float64)
    for i in range(len(t)):
        dd[i]=p[i]/287.05/(t[i]+273.15)
    return dd

ex=en(dfx['t'].values,dfx['sp'].values)

output:

array([1.15764165, 1.15852414, 1.15810831, 1.1579697 , 1.15822752,
       1.15864432, 1.15906853, 1.15852962, 1.15669754, 1.15534144])

That will fix it!

answered Dec 22, 2019 at 19:49

oppressionslayer

7,2242 gold badges11 silver badges26 bronze badges

Sign up to request clarification or add additional context in comments.

3 Comments

user2727167 Over a year ago

Unfortunately, that doesn't work on my machine. I get same error. I tried passing directly with dd = np.empty((dfx['t'].values.size),np.float64), i get error TypingError: cannot determine Numba type of <class 'pandas.core.frame.DataFrame'>

oppressionslayer Over a year ago

@user2727167, interesting. float16 definitely didnt work on my machine, it gave the same error you got, but the other two worked. What versions of the packages are your running, maybe you have an older version that just needs to be pip upgraded. just trying to think out of the box here, since the higher floats worked on mine. I have numpy '1.16.2', pandas 0.25.3, and numba '0.46.0'

user2727167 Over a year ago

I found out what the problem was. List of values in dfx['t'] was defined as float16 in original dataframe and even had some NaNs. When filtering this everything worked fine. Thank you.

Collectives™ on Stack Overflow

Numba won't accept numpy array datatype

1 Answer 1

3 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

3 Comments

Your Answer

Sign up or log in

Post as a guest

Related