새싹 AI데이터엔지니어 핀테커스 5주차 (금) - Boxplot, Numpy

❤️ 배운 것

Boxplot

IQR = Q3- Q1
inter quartile range

whis => default 1.5
위스커(박스플랏의 수염: 위아래) 바깥의 점은 outlier

def box_plot_test():  
    n_student = 100  
    math_scores = np.random.normal(loc=50, scale=10, size=(100,))  

    fig, ax = plt.subplots(figsize=(7, 7))  
    ax.boxplot(math_scores)

 

showfliers -> outlier 보여줄지 여부
vert -> false면 수평으로 보여주기

mean: 평균
median: 중앙값

median_props는 중앙값 보여주는 설정
k: black, b: blue

Box plot variation

def box_plot_test():  
    n_student = 100  
    math_scores = np.random.normal(loc=50, scale=10, size=(100,))  

    fig, axes = plt.subplots(nrows=3, ncols=4, figsize=(20, 15))  
    # axes = axes.flatten()  

    ''' basics '''  
    axes[0, 0].set_title('basic')  
    axes[0, 0].boxplot(math_scores)  
    axes[0, 1].set_title('notch')  
    axes[0, 1].boxplot(math_scores, notch=True)  
    axes[0, 2].set_title('whisker')  
    axes[0, 2].boxplot(math_scores, notch=True, whis=2)  
    axes[0, 3].set_title('sym')  
    axes[0, 3].boxplot(math_scores, notch=True, whis=1, sym='bx')  

    '''showfliers, vert, median props'''  
    axes[1, 0].set_title('showfliers')  
    axes[1, 0].boxplot(math_scores, notch=True, showfliers=False)  
    axes[1, 1].set_title('vert')  
    axes[1, 1].boxplot(math_scores, notch=True, showfliers=False, vert=False)  

    median_props = {'linewidth': 2, 'color': 'k'}  
    axes[1, 2].set_title('median_props')  
    axes[1, 2].boxplot(math_scores, medianprops=median_props)  

    box_props = {'linestyle': '--', 'color': 'k', 'alpha': 0.7}  
    axes[1, 3].set_title("box props")  
    axes[1, 3].boxplot(math_scores, medianprops=median_props, boxprops=box_props)  

    whisker_props = {'linestyle': '--', 'color': 'tab:blue', 'alpha': 0.8}  
    axes[2, 0].set_title("whiskerprops, capprops")  
    axes[2, 0].boxplot(math_scores, medianprops=median_props, boxprops=box_props,  
                       whiskerprops=whisker_props, capprops=whisker_props)  


    ''' hstacked data '''  
    n_student = 100  
    math_scores = np.random.normal(loc=50, scale=15, size=(100, 1))  
    chem_scores = np.random.normal(loc=70, scale=10, size=(n_student, 1))  
    phy_scores = np.random.normal(loc=30, scale=12, size=(n_student, 1))  
    pro_scores = np.random.normal(loc=80, scale=5, size=(n_student, 1))  

    data = np.hstack((math_scores, chem_scores, phy_scores, pro_scores))  
    axes[2, 1].set_ylim([0, 100])  
    axes[2, 1].set_title("hstacked data")  
    axes[2, 1].boxplot(data)  

    ''' major minor tick '''  
    axes[2, 2].set_title("major minor tick")  
    # n_student = 100  
    math_scores = np.random.normal(loc=50, scale=15, size=(100, 1))  
    english_scores = np.random.normal(loc=70, scale=10, size=(100, 1))  
    physics_scores = np.random.normal(loc=30, scale=12, size=(100, 1))  
    programming_scores = np.random.normal(loc=80, scale=5, size=(100, 1))  
    data = np.hstack((math_scores, english_scores, physics_scores, programming_scores))  
    median_props2 = {'linewidth': 1, 'color': 'tab:red'}  
    box_props2 = {'linewidth': 1.5, 'color': 'k', 'alpha': 0.7}  
    whisker_props2 = {'linestyle': '--', 'color': 'tab:blue'}  

    # labels  
    labels = ['Math', 'English', 'Physics', 'Programming']  
    axes[2, 2].set_xticklabels(labels)  
    # axes[2, 2].tick_params(labelsize=20)  
    axes[2, 2].tick_params(axis='x', rotation=10)  

    axes[2, 2].boxplot(data, labels=labels,  
                       notch=True, medianprops=median_props2, boxprops=box_props2,  
                       whiskerprops=whisker_props2, capprops=whisker_props2)  

    axes[2, 2].set_ylim([0, 100])  

    major_yticks = np.arange(0, 101, 20)  
    minor_yticks = np.arange(0, 101, 5)  

    axes[2, 2].set_yticks(major_yticks)  
    axes[2, 2].set_yticks(minor_yticks, minor=True)  

    axes[2, 2].grid(axis='y', linewidth=2)  
    axes[2, 2].grid(axis='y', which='minor', linewidth=2, linestyle=':')  
    axes[2, 2].grid(axis='x', linewidth=0)  

    ''' violin plot '''  
    n_group = np.arange(len(labels))  
    axes[2, 3].set_title("violin plot")  
    axes[2, 3].violinplot(data, positions=n_group)  
    axes[2, 3].set_xticks(n_group)  
    axes[2, 3].set_xticklabels(labels)  
    axes[2, 3].tick_params(axis='x', rotation=10)  

    # axes[2, 2].tick_params(labelsize=20, bottom=False, labelbottom=False) # 라벨 지우기  
    # fig.subplots_adjust(hspace=0.1) # horizontal space(위아래)를 10%로 줄이는 것

 

Numpy

ndarray 기본 생성 방법

vec_np = np.array([1, 2, 3])  

np.array() 안에 list를 넣으면 ndarray가 생성됨

np.full((2, 2), np.inf)
array([[inf, inf],
       [inf, inf]])
>>> np.full((2, 2), 10)
array([[10, 10],
       [10, 10]])

• np.inf = $\infty$

arange

np.arange(10)
Out[8]: array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
np.arange(2, 5)
Out[9]: array([2 3 4])
np.arange(2, 10, 2)
Out[10]: array([2, 4, 6, 8])
np.arange(1.5, 10.5)
Out[11]: array([1.5, 2.5, 3.5, 4.5, 5.5, 6.5, 7.5, 8.5, 9.5])
np.arange(1.5, 10.5, 2.5)
Out[12]: array([1.5, 4. , 6.5, 9. ])

• np.arange는 1:시작, 2:끝, 3:스텝(구간)으로 생성

linspace

np.linspace(0, 1, 5)
Out[13]: array([0.  , 0.25, 0.5 , 0.75, 1.  ])
np.linspace(0, 1, 10)
Out[14]: 
array([0.        , 0.11111111, 0.22222222, 0.33333333, 0.44444444,
       0.55555556, 0.66666667, 0.77777778, 0.88888889, 1.        ])

• np.linspace는 1:시작, 2:끝, 3:개수로 생성

randn

random_values = np.random.randn(300)
'''
[0.         0.11111111 0.22222222 0.33333333 0.44444444 0.55555556
 0.66666667 0.77777778 0.88888889 1.        ]
(300,)
'''

np.random.normal

normal = np.random.normal(loc=[-2, 0, 3],  
                          scale=[1, 2, 5],  
                          size=(200, 3))

• 200 x 3 크기의 정규분포에서 랜덤 선택한 값 3묶음

단일 평균, 표준편차로 nxn 추출

normal = np.random.normal(loc=-2, scale=1, size=(3, 3))  
print(normal)
'''
[[-0.21198265 -2.51857286 -2.0761132 ]
 [-2.16084686 -1.50902093 -1.63418275]
 [-1.80634427 -2.61567043 -2.04913825]]
'''

randn vs rand

random_values = np.random.randn(300)
uniform = np.random.rand(1000)

• randn : 표준정규분포로 평균 0 분산 1 기준으로 300개 추출
• rand : uniform 무작위로 0~1 사이 값 1000개 추출

normal vs uniform

normal = np.random.normal(loc=[-2, 0, 3],  
                          scale=[1, 2, 5],  
                          size=(200, 3))
uniform = np.random.uniform(low=-10, high=10, size=(10000, ))

• normal : 정규분포로 평균, 분산을 정해서 200개 추출
• uniform : 무작위로 low ~ high 사이 값 10000개 추출

randint

randint = np.random.randint(low=0, high=7, size=(20,))

• randint : low에서 high 사이의 정수 20개 추출

size

M = np.ones(shape=(10,))  
N = np.ones(shape=(3, 4))  
O = np.ones(shape=(3, 4, 5))  
P = np.ones(shape=(2, 3, 4, 5, 6))  

print("size of M: ", M.size)  # size of M:  10
print("size of N: ", N.size)  # size of N:  12
print("size of O: ", O.size)  # size of O:  60
print("size of P: ", P.size)  # size of P:  720

• size : 원소의 개수

reshape

a = np.arange(6)  
b = np.reshape(a, (2, 3))  

print("original ndarray: \n", a)  
print("reshaped ndarray: \n", b)
'''
original ndarray: 
 [0 1 2 3 4 5]
reshaped ndarray: 
 [[0 1 2]
 [3 4 5]]
'''

a = np.arange(24)  
b = np.reshape(a, (2, 3, 4))

print("original ndarray: \n", a)  
print("reshaped ndarray: \n", b)
'''
original ndarray: 
 [ 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23]
reshaped ndarray: 
 [[[ 0  1  2  3]
  [ 4  5  6  7]
  [ 8  9 10 11]]

 [[12 13 14 15]
  [16 17 18 19]
  [20 21 22 23]]]
'''

a = np.arange(12)  

b = a.reshape((2, -1))  
c = a.reshape((3, -1))  
d = a.reshape((4, -1))  
e = a.reshape((6, -1))

print(b.shape, c.shape, d.shape, e.shape)
''' (2, 6) (3, 4) (4, 3) (6, 2) '''

a = np.random.randint(0, 10, size=(2, 2))  
print(a)  
'''
[[8 2]
 [0 5]]
'''

row_vector = a.reshape(1, -1)  
col_vector = a.reshape(-1, 1)  
print(row_vector.shape, col_vector.shape)
''' (1, 4) (4, 1) '''

# flatten : vector로 만들어 주는 것  
M = np.arange(9)  
N = M.reshape((3, 3))  
O = N.flatten()  

print(M, '\n')  
print(N, '\n')  
print(O, '\n')  
print(M.shape, O.shape)
'''
[0 1 2 3 4 5 6 7 8] 

[[0 1 2]
 [3 4 5]
 [6 7 8]] 

[0 1 2 3 4 5 6 7 8] 
'''

broadcasting

# 차원이 같은 경우  
A = np.arange(9).reshape(3, 3)  
B = 10 * np.arange(3).reshape((-1, 3))  
C = A + B  

A = np.arange(3).reshape((3, -1))  
B = 10 * np.arange(3).reshape((-1, 3))  
C = A + B  

print("A: {}/{}\n{}".format(A.ndim, A.shape, A))  
print("B: {}/{}\n{}\n".format(B.ndim, B.shape, B))  

print("A + B: {}/{}\n{}".format(A.ndim, C.shape, C))

'''
A: 2/(3, 1)
[[0]
 [1]
 [2]]
B: 2/(1, 3)
[[ 0 10 20]]
'''

# (2,3,3) + (1,3,3) = (2,3,3)
A = np.arange(2*3*3).reshape((2, 3, 3))  
B = 10*np.arange(3*3).reshape((1, 3, 3))  

C = A + B  

print("A: ", A)  
print("B: ", B)  
print("C: ", C)

'''
A:  [[[ 0  1  2]
  [ 3  4  5]
  [ 6  7  8]]

 [[ 9 10 11]
  [12 13 14]
  [15 16 17]]]
B:  [[[ 0 10 20]
  [30 40 50]
  [60 70 80]]]
C:  [[[ 0 11 22]
  [33 44 55]
  [66 77 88]]

 [[ 9 20 31]
  [42 53 64]
  [75 86 97]]]
'''

 

 

2차원을 3차원으로 broadcasting

def ndarray_test5():  
    A = np.arange(2*3*4).reshape((2, 3, 4))  
    B = 10*np.arange(3*4).reshape((3, 4))  
    C = A + B  

    print(C)
'''
array([[[  0,  11,  22,  33],
        [ 44,  55,  66,  77],
        [ 88,  99, 110, 121]],
       [[ 12,  23,  34,  45],
        [ 56,  67,  78,  89],
        [100, 111, 122, 133]]])
'''

 

 

+ ( 3 x 1 ) 도 ( 3 x 4 ) 로 broadcasting이 가능해서 ( 2 x 3 x 4 ) 와 broadcasting 연산 가능

 

ndarray indexing

def ndarray_test6():  
    a = np.arange(10)  
    print(f"ndarray: \n{a}")  
    # [0 1 2 3 4 5 6 7 8 9]  
    print(f"a[:3]", a[:3])  
    # a[:3] [0 1 2]  
    print(f"a[::3]: ", a[::3])  
    # a[::3]:  [0 3 6 9]  

    a = np.arange(9).reshape((3, 3))  
    print(f"ndarray: \n{a}")  
    # [[0 1 2]  
    #  [3 4 5]    #  [6 7 8]]    print(a[0][1])  
    # 1  

    a = np.arange(12).reshape((4, 3))  
    print(f"ndarray: \n{a}")  
    # [[ 0  1  2]  
    #  [ 3  4  5]    #  [ 6  7  8]    #  [ 9 10 11]]    print("a[1:3, 2]", a[1:3, 2])  
    # a[1:3, 2] [5 8]

 

 

💛 배운점/느낀점

- numpy로 random 추출하는 방법을 익힘, reshape 및 tensor 계산하는 방법도 익힘

- numpy broadcasting 원리를 배움

- 자습: 선형대수와 통계학으로 배우는 머신러닝 with 파이썬의 마지막 챕터 딥러닝 학습 중

- 다음 주 부터는 학원에서 머신러닝을 배울 예정