liuyubobobo《机器学习》学习笔记（五）

numpy和matplotlib (2)

---

1.矩阵运算

• 运算符：+、-、*、/、//、**、%……

• abs()

• sin() cos() tan()

• exp() log()

• power() log2() log10()

• dot() 矩阵乘法

  A * B 只是矩阵各位置相乘，A.dot(B) 才是矩阵乘法

• A.T 转置

• inv() 逆矩阵

  A
  # array([[0, 1],
  #        [2, 3]])
  np.linalg.inv(A)
  # array([[-1.5,  0.5],
  #        [ 1. ,  0. ]])
  

• pinv() 伪逆矩阵 （不要求是方阵）

  C
  # array([[0, 1, 2],
  #        [3, 4, 5]])
  D = np.linalg.pinv(C)
  # array([[-0.77777778,  0.27777778],
  #        [-0.11111111,  0.11111111],
  #        [ 0.55555556, -0.05555556]])
  C.dot(D)
  # array([[ 1.00000000e+00, -4.16333634e-17],
  #        [ 5.55111512e-16,  1.00000000e+00]])
  

2.聚合运算

聚合：把一组值变成一个值

• sum()

  X = np.arange(16).reshape((4, 4))
  np.sum(X)
  # 120
  np.sum(X, axis=0)
  # array([24, 28, 32, 36])
  np.sum(X, axis=1)
  # array([ 6, 22, 38, 54])
  

• max() min()

• prod() 求积

• mean() 均值

• median() 中位数

• percentile() 百分位数

  L = np.array(np.random.random((10, 10)))
  np.percentile(L, q = 50)
  # 0.4706891049878974
  for percent in [0, 25, 50, 75, 100]:
      print(np.percentile(L, q = percent))
  # 0.0013510975813705306
  # 0.18301899481429829
  # 0.4706891049878974
  # 0.7229884292204094
  # 0.9879405960925149
  

• var()方差 std()标准差

3.arg运算

• argmin() argmax() 最小（大）值的索引

  x = np.random.normal(0, 1, size = 1000)
  np.min(x)
  # -2.949169178250681
  np.argmin(x)
  # 42
  x[42]
  # -2.949169178250681
  

• sort()

  X
  # array([[5, 8, 9, 3],
  #        [5, 5, 4, 8],
  #        [3, 1, 8, 5],
  #        [9, 3, 4, 7]])
  np.sort(X)   # 同np.sort(X, axis=1)
  # array([[3, 5, 8, 9],
  #        [4, 5, 5, 8],
  #        [1, 3, 5, 8],
  #        [3, 4, 7, 9]])
  np.sort(X, axis=0)
  # array([[3, 1, 4, 3],
  #        [5, 3, 4, 5],
  #        [5, 5, 8, 7],
  #        [9, 8, 9, 8]])
  

• argsort() 排序后每个数在原数组中的索引

  x
  # array([ 4, 10, 14,  8,  9, 12,  1,  6,  1,  6, 11,  3, 10, 10, 10,  2])
  np.argsort(x) # 索引为6的数排到第一位
  # array([ 6,  8, 15, 11,  0,  7,  9,  3,  4,  1, 12, 13, 14, 10,  5,  2],
  #       dtype=int32)
  

• partition()

  np.partition(x, 6)    # 小于6的排在6前面，大于6的排在6后面
  # array([ 1,  2,  3,  1,  4,  6,  6,  8,  9, 10, 10, 10, 10, 11, 14, 12])
  

• argpartition()

  np.argpartition(x, 6)
  # array([ 6, 15, 11,  8,  0,  9,  7,  3,  4, 14, 13,  1, 12, 10,  2,  5],
  #       dtype=int32)
  

4.Fancy indexing

x = np.arange(16)
# 取索引为3，5，8的值可以
[x[3], x[5], x[8]]
# 也可以
ind = [3, 5, 8]
x[ind]
ind = np.array([[0, 2],
                [1, 3]])
x[ind]

# 矩阵类似
X = x.reshape(4, -1)
row = np.array([0, 1, 2])
col = np.array([1, 2, 3])
X[row, col]           # 取(0, 1), (1, 2), (2, 3)三个点
# array([ 1,  6, 11])

# 另一种方法
col = [True, False, True, True]
X[1:3, col]
# array([[ 4,  6,  7],
#        [ 8, 10, 11]])