NumPy.md

NumPy

import numpy as np

Modifying ndarray dimensions

Use np.tile() to add extra dimensions. Use np.moveaxis() to exchange dimensions.

>>> A = np.ones([2, 3])
>>> B = np.tile(A, [4, 5, 1, 1])
>>> B.shape
(4, 5, 2, 3)
>>> C = np.moveaxis(B, [0, 1, 2, 3], [2, 3, 0, 1])
>>> C.shape
(2, 3, 4, 5)

Slicing ndarrays without losing dimensions

The following index slice loses a dimension from the ndarray:

>>> a = np.ones([5, 4, 1])
>>> b = a[:, 0, :]
>>> b.shape
(4, 1)

To keep the same dimension structure, use this instead:

>>> c = a[:, [0], :]
>>> c.shape
(4, 1, 1)

This might occur if assigning one array to another with a variable dimension size which may be 1.

Random numbers

See https://numpy.org/doc/stable/reference/random/index.html?highlight=random#module-numpy.random for updated methods

from numpy.random import default_rng
rng = default_rng(5)
vals = rng.standard_normal(10) # standard deviation with shape (10,)
vals = rng.random((10, 2, 4)) # uniform distribution with shape (10, 2, 4)

default_rng(n) initialized the seed using the integer n, ensuring reproducability. If left blank, will initialize with random seed.

Save/load data (csv)

If want to read in other programmes and np.ndarray has at most two dimensions, can save as csv:

data = np.array([1, 2, 3, 4])
np.savetxt('data.csv', data, delimiter=',') # save
data = np.loadtxt('data.csv', delimiter=',') # load

Save/load data (npy)

If np.ndarray has more than two dimensions and will only be loading in Python, use np.save():

data = np.ones([3, 4, 5])
np.save('data.npy', data)
data = np.load('data.npy')