Numpy Cheat Sheet

Basics 基础

简单创建

np.arange(start, stop, step)
np.array([1, 2, 3])

Placeholders 创建

包括创建 linespace, zeros, ones, random, empty 数组

np.linespace(start, stop, num)
np.zeros(shape)
np.ones(shape)
np.random.random(shape)	# 0-1 uniform
np.random.randn(*dimensions)	# normal/gaussian
np.random.randint(low, high, shape)	# high is exclusive
np.empty(shape)

Array 数组

Array Properties 属性

array.shape
array.ndim	# dimensions of array
array.size
array.dtype
array.astype(type)	# convert data type
len(array)
type(array)

Copy 复制

np.copy(array)
other = array.copy()

Sort 排序

array.sort()
array.sort(axis=0)

Array Operations 数组操作

Adding & removing

np.append(a, b)
np.insert()	# not common
np.delete()	# not common

Combining

np.concatenate(arrays, axis)
np.vstack(arrays)
np.hstack(arrays)
np.stack(arrays, axis)	# This will create a new axis

Shaping

array.reshape(shape)
array.flatten()
array.transpose()	# equals array.T
array.transpose(axes)	# permute axes

Math 数学

Operations 基础运算

# basic
np.multiply(x, y)	# equals x @ y
np.dop(x, y)		# dot product of 1D array
np.sqrt(x)
np.sin(x)
np.cos(x)
np.log(x)
np.exp(x)
np.power(x1, x2)	# x1 & x2 have the same shape or x2 can broadcast
np.ceil(x)
np.floor(x)

# preprocess
np.isnan(x)
np.isinf(x)
np.round(x, ndigits)
np.nan_to_num(x, nan)	# Replace NaN with zero and infinity with large finite numbers (default behaviour) or with the numbers defined by the user using the nan, posinf and/or neginf keywords.
np.all(x, axis)		# test whether all be true along a axis
np.any(x, axis)

# compare
np.max(x, axis)
np.min(x, axis)
np.maximum(x, y)	# Element-wise maximum of array elements.
np.minimum(x, y)

# cumulative
np.cumsum(x, axis)	# cumulative sum, if axis=None, flatten x
np.diff(x, axis)	# differences
np.prod(x, axis)	# product along given axis

# arg-relative
np.argmax(x, axis)
np.argsort(x, axis)
np.bincount()		# Count number of occurrences of each value in array of non-negative ints.

# linear algebra
np.linalg.det(x)
np.linalg.inv(x)

# Statistics
np.sum(x, axis)
np.mean(x, axis)
np.std(x, axis)
np.corrcoef(x, y=None)

Slicing 切片

# n dimensions
array[:3,:,...]	# upper bound is exclusive
array[:,-1]		# reverse slicing
array[::2,:]	# step is 2

# bool
array[array > 5]
array[(array>5) & (array%2==0)]

# Fancy indexing
array[[1, 2, 3],:]	# could be any iterable int array

Broadcast 广播

广播数组维度需要满足以下要求任意一个：

1. 从后往前比，两个数组各个维度大小相同

   A = np.zeros((2, 3, 4, 5))
   B = np.zeros((4, 5))
   C = A + B	# B will broadcast to (2, 3, 4, 5)

2. 两个数组存在维度大小不相等时，其中一个不相等维度大小为1

   A = np.zeros((2, 3, 4, 5))
   B = np.zeros((4, 1))
   C = np.zeros((2, 1, 1, 5))
   D = A + B + A * C	# B & C will broadcast to (2, 3, 4, 5)

存储为 .bin

使用 tofile & fromfile 将 ndarray 存储为二进制文件

import numpy as np
from pathlib import Path

bin_file = Path('./xxx.bin')
a = np.array([1, 2, 3, 4])
with open(binfile, mode='w') as f:
    a.tofile(f)
b = np.fromfile(binfile, dtype=a.dtype)
print(b)