numpy.md

save and load

• save and load as npy, one array at a time

# .npy will be automatically added, i.e., test.npy
np.save('test', data)
data = np.load('test.npy')

• save and load as npz, multiple array

np.save('test', val1, mykey=val2)

# data will be like a dict, key would be 'arr_0', 'arr_1' if not specified
data = np.load('test.npz')

npz to csv

for npz_file in os.listdir(dir):
    if npz_file.endswith('.npz'):
       # data works like a dictionary
        data = np.load(os.path.join(dir, npz_file))
        csv_file = os.path.join(dir, npz_file[:-4] + '.csv')
        np.savetxt(csv_file, data['embedding'], delimiter=',')

prevent divide by zero

data = np.divide(data-mu, sigma, out=np.zeros_like(data), where=sigma!=0)

indexing and slicing

• slice will get a veiw of the original, i.e., a reference, wont copy
• basic indexing

# basic syntex, obj can be slice, integer, a tuple of slice or integer
x[obj]

# start:end:step slice
x[1:7:2]

• if dim of obj is less than N, then : is assumed
• Ellipsis expands to the number of : objects needed for the selection tuple to index all dimensions.

>>> x = np.array([[[1],[2],[3]], [[4],[5],[6]]])
>>> x[...,0]
array([[1, 2, 3],
       [4, 5, 6]])

• np.newaxis add new axis, its a alias for 'None'

>>> x.shape
(2, 3, 1)
>>> x[:,np.newaxis,:,:].shape
(2, 1, 3, 1)

# or use None
x[:, None, :, :]

• ::

# slice every step elements
>>> a = [1,2,3,4,5,6,7,8,9]
>>> a[::3]
[1, 4, 7]

# specify start
>>> a[2::3]
[3, 6, 9]

np.where

• np.where(condition[,x, y])
  • Return elements chosen from x or y depending on condition.
• np.where(x)
  • same as np.nonzero(x)
  • return indices of the elements that are non-zero.