Numpy. section7 : API들의 axis, keepdims 인자. Lec28. 텐서 ndarray에서의 axis, keepdims

[1]

• ( 2, 3, 4 ) 를 keepdims를 안했을경우 ( 3, 4 )
  keepdims를 했을 경우 ( 1, 3, 4 )가 된다

a = np.arange(2*3*4).reshape((2,3,4))

sum_ = a.sum(axis = 0)

print("ndarray: {}\n{}".format(a.shape, a))
print("ndarray.sum(axis = 0): {}\n{}".format(sum_.shape, sum_))

• (2,3,4) -> (2,4) or (2,1,4)
  (2,4)일때는 브로드캐스팅 불가
  (2,1,4)일때는 브로드캐스팅 가능
  

a = np.arange(2*3*4).reshape((2,3,4))

sum_ = a.sum(axis = 1)
sum_k = a.sum(axis = 1, keepdims = True)

print("ndarray: {}\n{}".format(a.shape, a))
print("axis = 1: {}\n{}".format(sum_.shape, sum_))
print("axis = 1, keepdims = True: {}\n{}".format(sum_k.shape, sum_k))

a = np.arange(2*3*4).reshape((2,3,4))

sum_ = a.sum(axis =2)
sum_k = a.sum(axis = 2, keepdims = True)

print("ndarray: {}\n{}".format(a.shape, a))
print("axis = 2: {}\n{}".format(sum_.shape, sum_))
print("axis = 2, keepdims = True: {}\n{}".format(sum_k.shape, sum_k))

[2] 전체코드

import numpy as np

a = np.arange(2*3*4).reshape((2,3,4))

sum_ = a.sum(axis = 0)

print("ndarray: {}\n{}".format(a.shape, a))
print("ndarray.sum(axis = 0): {}\n{}".format(sum_.shape, sum_))

#

a = np.arange(2*3*4).reshape((2,3,4))

sum_ = a.sum(axis = 1)
sum_k = a.sum(axis = 1, keepdims = True)

print("ndarray: {}\n{}".format(a.shape, a))
print("axis = 1: {}\n{}".format(sum_.shape, sum_))
print("axis = 1, keepdims = True: {}\n{}".format(sum_k.shape, sum_k))

#

a = np.arange(2*3*4).reshape((2,3,4))

sum_ = a.sum(axis =2)
sum_k = a.sum(axis = 2, keepdims = True)

print("ndarray: {}\n{}".format(a.shape, a))
print("axis = 2: {}\n{}".format(sum_.shape, sum_))
print("axis = 2, keepdims = True: {}\n{}".format(sum_k.shape, sum_k))

#

n_test_time, n_student, n_class = 4, 3, 4
m_score, M_score = 0, 100 
scores = np.random.randint(low = m_score, high = M_score, size = (n_test_time, n_student, n_class))
print("scores: \n", scores)

score_mean = np.mean(scores, axis = 0)
print("score mean: ", score_mean.shape, '\n', score_mean)

score_mean = np.mean(scores, axis = 1)
print("score mean: ", score_mean.shape, '\n', score_mean)

C, H, W = 3, 100, 200

# (C, H, W) case
images = np.random.randint(0, 256, size = (C, H, W))

print("Shape of original image:", images.shape)

gray_image = np.mean(images, axis = 0)
print("Shape of gray-scaled image:", gray_image.shape, '\n')

# (H, W, C) case
images = np.random.randint(0, 256, size = (H, W, C))

print("Shape of original image:", images.shape)

gray_image = np.mean(images, axis = -1)
print("Shape of gray-scaled image:", gray_image.shape, '\n')