Question

I am trying to optimize this code using numpy built-in function:

result = [test[0]]
for x in range(1, len(test)):
    if test[x] <= result[-1]:
        result.append(result[-1]+1)
    else:
        result.append(test[x])
print(result)

The above code loops through the array, checks if the previous value is equal or superior to the current value, and if it is, adds +1. Otherwise, it does nothing.

It is recursive (the current value depends on the previously calculated value). My input array is, by design, always sorted (in ascending order).

∗∗∗∗∗ 我预期会收到<条码>[0、1、2、3、4、15、16、17、18、19]。

是否有更好的办法这样做? 我需要多次使用非常大的阵列(>=10M长)。

Answer 1

如何?

import numpy as np

def adjust_array(test):
    result = [test[0]]
    for i in range(1, len(test)):
        result.append(max(test[i], result[-1] + 1))
    return np.array(result)

# Test the function
test = np.array([0, 0, 0, 1, 4, 15, 16, 16, 16, 17])
result = adjust_array(test)
print(result)

这确保,如果投入阵列中的相应内容不大,结果阵列中的每个要素都严格高于前一个。

产出:

>>> [ 0  1  2  3  4 15 16 17 18 19]

Answer 2

我在《诺姆布评论》之后测试了不同的方法:

import numpy as np
import pandas as pd
from timeit import timeit
import itertools


# Original loop-based version
def adjust_array_if(test):
    result = [test[0]]
    for i in range(1, len(test)):
        if test[i] <= result[-1]:
            result.append(result[-1]+1)
        else:
            result.append(test[i])
    return np.array(result)

# Max loop-based version
def adjust_array_max(test):
    result = [test[0]]
    for i in range(1, len(test)):
        result.append(max(test[i], result[-1] + 1))
    return np.array(result)

# Version using itertools
def adjust_array_itertools(test):
    result = np.fromiter(itertools.accumulate(test, lambda x, y: max(x + 1, y)), dtype=np.int32)
    return result

# Set up the DataFrame
functions = pd.Index([ adjust_array_if ,  adjust_array_max ,  adjust_array_itertools ], name= function )
lengths = np.arange(100, 11000, 1000)
results = pd.DataFrame(index=lengths, columns=functions)

# Time the functions
for i in lengths:
    a = np.random.randint(1, i, size=i)
    for j in functions:
        stmt = f"{j}(a)"
        setup = f"from __main__ import {j}, a"
        timing = timeit(stmt, setup, number=10)
        results.at[i, j] = timing

# Plot the results
results.plot()

with the following results:

Simpler is better in that case?

友情链接