Numbaのjitで行列の計算を早くしたい

Question

numbaのjitを用いて行列の計算を早くしたい。

pythonで関数を定義してjitを使って計算速度を早くしようとしたら次のようなエラーがでました。
このエラーはどうやったら取り除けるのでしょうか？

発生している問題・エラーメッセージ

sum.py:13: NumbaWarning: 
Compilation is falling back to object mode WITH looplifting enabled because Function "LOOP" failed type inference due to: Invalid use of Function(<built-in function setitem>) with argument(s) of type(s): (readonly array(float64, 2d, C), UniTuple(int64 x 2), float64)
 * parameterized
In definition 0:
    All templates rejected with literals.
In definition 1:
    All templates rejected without literals.
In definition 2:
    All templates rejected with literals.
In definition 3:
    All templates rejected without literals.
In definition 4:
    All templates rejected with literals.
In definition 5:
    All templates rejected without literals.
In definition 6:
    All templates rejected with literals.
In definition 7:
    All templates rejected without literals.
In definition 8:
    TypeError: Cannot modify value of type readonly array(float64, 2d, C)
    raised from /usr/local/lib/python3.7/site-packages/numba/typing/arraydecl.py:179
In definition 9:
    TypeError: Cannot modify value of type readonly array(float64, 2d, C)
    raised from /usr/local/lib/python3.7/site-packages/numba/typing/arraydecl.py:179
This error is usually caused by passing an argument of a type that is unsupported by the named function.
[1] During: typing of setitem at sum.py (19)

File "sum.py", line 19:
def LOOP():
    <source elided>
                for l in range(D1):
                    o[i,j] += i+j+k+l
                    ^

  @jit
sum.py:13: NumbaWarning: 
Compilation is falling back to object mode WITHOUT looplifting enabled because Function "LOOP" failed type inference due to: cannot determine Numba type of <class 'numba.dispatcher.LiftedLoop'>

File "sum.py", line 15:
def LOOP():
    for i in range(D1):
    ^

  @jit
/usr/local/lib/python3.7/site-packages/numba/object_mode_passes.py:178: NumbaWarning: Function "LOOP" was compiled in object mode without forceobj=True, but has lifted loops.

File "sum.py", line 15:
def LOOP():
    for i in range(D1):
    ^

  state.func_ir.loc))
/usr/local/lib/python3.7/site-packages/numba/object_mode_passes.py:188: NumbaDeprecationWarning: 
Fall-back from the nopython compilation path to the object mode compilation path has been detected, this is deprecated behaviour.

For more information visit http://numba.pydata.org/numba-doc/latest/reference/deprecation.html#deprecation-of-object-mode-fall-back-behaviour-when-using-jit

File "sum.py", line 15:
def LOOP():
    for i in range(D1):
    ^

  state.func_ir.loc))
sum.py:13: NumbaWarning: 
Compilation is falling back to object mode WITHOUT looplifting enabled because Function "LOOP" failed type inference due to: Invalid use of Function(<built-in function setitem>) with argument(s) of type(s): (readonly array(float64, 2d, C), UniTuple(int64 x 2), float64)
 * parameterized
In definition 0:
    All templates rejected with literals.
In definition 1:
    All templates rejected without literals.
In definition 2:
    All templates rejected with literals.
In definition 3:
    All templates rejected without literals.
In definition 4:
    All templates rejected with literals.
In definition 5:
    All templates rejected without literals.
In definition 6:
    All templates rejected with literals.
In definition 7:
    All templates rejected without literals.
In definition 8:
    TypeError: Cannot modify value of type readonly array(float64, 2d, C)
    raised from /usr/local/lib/python3.7/site-packages/numba/typing/arraydecl.py:179
In definition 9:
    TypeError: Cannot modify value of type readonly array(float64, 2d, C)
    raised from /usr/local/lib/python3.7/site-packages/numba/typing/arraydecl.py:179
This error is usually caused by passing an argument of a type that is unsupported by the named function.
[1] During: typing of setitem at sum.py (19)

File "sum.py", line 19:
def LOOP():
    <source elided>
                for l in range(D1):
                    o[i,j] += i+j+k+l
                    ^

  @jit
/usr/local/lib/python3.7/site-packages/numba/object_mode_passes.py:178: NumbaWarning: Function "LOOP" was compiled in object mode without forceobj=True.

File "sum.py", line 15:
def LOOP():
    for i in range(D1):
    ^

  state.func_ir.loc))
/usr/local/lib/python3.7/site-packages/numba/object_mode_passes.py:188: NumbaDeprecationWarning: 
Fall-back from the nopython compilation path to the object mode compilation path has been detected, this is deprecated behaviour.

For more information visit http://numba.pydata.org/numba-doc/latest/reference/deprecation.html#deprecation-of-object-mode-fall-back-behaviour-when-using-jit

File "sum.py", line 15:
def LOOP():
    for i in range(D1):
    ^

  state.func_ir.loc))

該当のソースコード

#coding:utf-8
from numba import jit
import math
import numpy as np
import matplotlib.pyplot as plt
import datetime
import time
now = datetime.datetime.now()
start = time.time()


D1 = 50
@jit
def LOOP():
    for i in range(D1):
        for j in range(D1):
            for k in range(D1):
                for l in range(D1):
                    o[i,j] += i+j+k+l
    return o

o= np.zeros((D1,D1))
A = list()
for p in range(3):
    if p<2:
        B = list(o)
        A.append(np.trace(LOOP()))
        o= np.zeros((D1,D1))
    else:
        B = list(o)
        A.append(np.trace(LOOP()))
        o= np.zeros((D1,D1))

print(A)

elapsed_time = time.time() - start
print ("elapsed_time:{0}".format(elapsed_time) + "[sec]")

Answer 1

リファレンスにもありますが、Numbaでは外部変数(グローバル変数)は、コンパイル時に定数として固定されてしまいます。
なのでreadonlyといったエラーが出ています。

解決策としては、ローカル変数になるようにすればいいです。→ 関数の引数で渡す。

import time
import numpy as np
from numba import jit

start = time.time()

@jit
def loop2(o, d):
    for i in range(d):
        for j in range(d):
            for k in range(d):
                for l in range(d):
                    o[i,j] += i+j+k+l
    return o

D1 = 50
A = list()
for p in range(3):
    o = np.zeros((D1,D1))
    if p < 2:
        A.append(np.trace(loop2(o, D1)))
    else:
        A.append(np.trace(loop2(o, D1)))

print(A)

elapsed_time = time.time() - start
print ("elapsed_time:{0}".format(elapsed_time) + "[sec]")

Comments

[textensor]

なるほどローカル変数じゃないとだめなのですね。
ありがとうございました。