Skip to content

Latest commit

 

History

History

numpy

Folders and files

NameName
Last commit message
Last commit date

parent directory

..

tools

tools

 
 

.DS_Store

.DS_Store

 
 

readme.md

readme.md

 
 

readme.md

常用

  • 修改数据类型
x = np.array(x,dtype=np.int32)
  • one-hot
num_classes = 10
arr = [1,3,4,5,9]
arr = np.eye(10)[arr]
  • 合并
x1 = np.array([[1,2]])
x2 = np.array([[3,4]])
np.concatenate((x1,x2),axis=0)  # 默认情况下，axis=0可以不写
>>array([[1, 2],
       [3, 4]])
  • 众数
a = np.array([1,2,3,1,2,1,1,1,3,2,2,1])
counts = np.bincount(a)
#bin的数量比x中的最大值大1，每个bin给出了它的索引值在x中出现的次数
#>>array([0, 6, 4, 2], dtype=int64)
np.argmax(counts)
#>> 1
  • 交换两行/列
#（例如opencv读取图片为RGB的顺序，想转位BGR）
img = cv2.imread("xx.jpg") #img.shape = h*w*3
img[:,:,[0,2]] = img[:,:,[2,0]]  #交换了第三维中的0列和2列 即BGR转为RGB
  • 求偏度、峰度
R = np.array([1, 2, 3, 4, 5, 6]) #初始化一组数据
R_mean = np.mean(R) #计算均值
R_var = np.var(R)  #计算方差
R_sc = np.mean(((R - R_mean)/R_var**0.5) ** 3) #计算偏斜度
R_ku = np.mean((R - R_mean) ** 4) / pow(R_var, 2) #计算峰度
print([R_mean, R_var, R_sc, R_ku])
  • 二维数组按列或者行求和
def non_zero_mean(np_arr):
    exist = (np_arr != 0)
    num = np_arr.sum(axis=1)
    #行为0列为1
    den = exist.sum(axis=1)
    return num/den
  • 生成位置矩阵（中间大四周小）
def positionMat(size):
    mat = np.zeros((size,size))
    even_size = size%2

    part_size = (size+1)//2
    mat_lt = np.ones((part_size, part_size))
    mat_lt = mat_lt * np.array(range(part_size)).reshape((part_size,1))
    mat_lt_triu = np.triu(mat_lt)
    mat_lt_triu2 = np.triu(mat_lt,k=1)#不复制对角线
    mat_lt = mat_lt_triu + mat_lt_triu2.T

    mat[:part_size, :part_size] = mat_lt
    mat_lb = np.flip(mat_lt,axis=0)
    mat[part_size:, :part_size] = mat_lb[even_size:,:]
    
    mat_rb = np.flip(mat_lb,axis=1)
    mat[part_size-even_size:, part_size:] = mat_rb[:,even_size:]

    mat_rt = np.flip(mat_rb,axis=0)
    mat[:part_size-even_size, part_size:] = mat_rt[:part_size-even_size,even_size:]
    return mat

size = 6
print(positionMat(size))
  • 元素级应用函数（类似pandas apply）
#https://blog.csdn.net/kudou1994/article/details/94417926
np.frompyfunc()
  • 多维数组索引
#比如 特征图f的shape=[8,1,4,4]，4x4代表特征图高宽   x=[0,3,2,1,0,0,0,0] y=[0,0,0,0,0,0,0,0]
#想取f的8个batchsize的对应x、y对应位置的值
#则需要每一个维度都构建对应的值
a0 = np.arange(8,dtype=np.int32).reshape(-1,1)
a1 = np.zeros(8,dtype=np.int32).reshape(-1,1)
x = x.reshape(-1,1)
y = y.reshape(-1,1)
res = f[a0,a1,x,y]
#res.shape = [8,1]
  • Sigmoid
def sigmoid(x):
    return 1 / (1 + np.exp(-x))
  • Softmax
def softmax(x, axis=None):
    x = x - x.max(axis=axis, keepdims=True)
    y = np.exp(x)
    return y / y.sum(axis=axis, keepdims=True)

工具