python-numpy-00

--------------------numpy-------------------setup--------

a = [ [1,2], [1,3] ]  #list

b = np.array(a, dtype=float)   #ndarray  -- array是一个函数 参数为一个python序列对象 

c = np.arange(start, end, step)

d = np.linspace(start, end, num)

np.random.randint(l,h,size=())

np.random.randn(r,c)     N(0,1)  #For random samples from , use: sigma*np.random.randn(...)+mu

random.shuffle()

def func(i, or i,j):  return i^2, or i*j

np.frommfuction(func, (3,)or (i,j) )

np.tile(ndarray,( i, j )   #堆砌ndarray(看出一个元素), i 行, j列 

numpy.vstack( (ndarray-a, ndarray-b) )

--------------------np.array.shape---------------------------

c = np.array()

c.shape = 3, -1

d = c.reshape( (3, -1))

-----------------------------std------average---------------------------------------------------

       
  r = np.random.binomial(1000, 0.8, 1000)   # 生成二项分布的随机数  
  std = np.std(r)   # 计算标准差
  avg = np.average(r)  # 计算平均数          
  np.max(x,0 or 1)  np.min(x,0 or 1)

      
  f = r[r>=(avg-std)]             # 选出落在一个平均值标准差范围内的元素
  f = f[f<=(avg+std)]   

-----------------------numpy.linalg-----------------

cos距离

norm = np.linalg.norm(sumx[1]) * np.linalg.norm(sumx[3]) #norm L2范数
dis = np.dot(sumx[1].transpose(), sumx[3]) / norm
dis = 0.5 + 0.5*dis          #归一化
print dis

欧式距离

dist = linalg.norm(A - B) 

sim = 1.0 / (1.0 + dist)  #归一化

------------------------------np.pad----------------------------------

np.pad(ndarr, ( (1,2), (2,1) ), mode)  # axis-i (begine,end)   mode- 'constant', 'edge' ...

np.pad(ndarray, (1,3), 'constant', constant_values=(0,0) )

http://docs.scipy.org/doc/numpy/reference/generated/numpy.pad.html

------------------save---load--------------

np.save('path-to-save.npy', ndarray)

ndarray = np.load('path')

np.savetxt('path-to-save.txt'., ndarray)  # 1 or 2

ndarray = np.loadxt('path')

---------------large-data-than-memory------------------

numpy.memmap( ) # similar to np.array( )

PyTables hdf5

-----------------random----------------------

np.random.shuffle(ndarray)

idx = np.random.permutation(n)  #np.range(n)

a[idx,:] ...

------------------------numpy--label---------broadcast----------------------------------

labels = [1 2 3 4 5 6] .... 

re = (numpy.arange(label_num) == labels[ : , None]).astype(numpy.float32)

--------------------reference---------------------

http://old.sebug.net/paper/books/scipydoc/numpy_intro.html

http://docs.scipy.org/doc/numpy/reference/generated/numpy.memmap.html

http://stackoverflow.com/questions/1053928/very-large-matrices-using-python-and-numpy