Numpy

http://numpy.org

High-performance vector, matrix and higher-dimensional data structures for Python. A core data type provider for the scientific python stack.

For a comprehensive tutorial, refer to this material:

http://wiki.scipy.org/Tentative_NumPy_Tutorial



In [3]:

    
import numpy as np

Q. What is an array in numpy? An compact representation of like-kind data.



In [4]:

    
array_data = [1, 2, 3, 4]



In [5]:

    
array = np.array(array_data)
array









    Out[5]:





array([1, 2, 3, 4])



In [6]:

    
array.dtype









    Out[6]:





dtype('int32')

Numpy has a large number of methods that translate to matlab/R/IDL commands. (ufuncs)



In [7]:

    
array.sum()









    Out[7]:





10



In [8]:

    
array.min(), array.mean(), array.max()









    Out[8]:





(1, 2.5, 4)



In [9]:

    
np.median(array)









    Out[9]:





2.5

Q. How do you index data quickly with numpy? Fancy indexing.



In [10]:

    
matrix = np.random.randint(0, 100, (5, 5))
matrix









    Out[10]:





array([[68, 30, 68, 51, 96],
       [92, 99, 79, 55, 92],
       [91, 55,  2, 22, 60],
       [24, 62,  5, 73, 44],
       [87, 23, 82, 16, 35]])



In [11]:

    
matrix.shape









    Out[11]:





(5, 5)



In [12]:

    
matrix > 50









    Out[12]:





array([[ True, False,  True,  True,  True],
       [ True,  True,  True,  True,  True],
       [ True,  True, False, False,  True],
       [False,  True, False,  True, False],
       [ True, False,  True, False, False]], dtype=bool)

Fancy indexing can be used to slice into data in complex ways (boolean expressions) and form views of data.



In [13]:

    
matrix[matrix>50]









    Out[13]:





array([68, 68, 51, 96, 92, 99, 79, 55, 92, 91, 55, 60, 62, 73, 87, 82])

Q. What if the coordinates are important? nonzero



In [14]:

    
rows, cols = np.nonzero(matrix>50)



In [15]:

    
matrix[rows, cols]









    Out[15]:





array([68, 68, 51, 96, 92, 99, 79, 55, 92, 91, 55, 60, 62, 73, 87, 82])

Broadcasting is a way to perform operations on data quickly avoiding iteration.



In [16]:

    
matrix = np.ones((3,3))
matrix









    Out[16]:





array([[ 1.,  1.,  1.],
       [ 1.,  1.,  1.],
       [ 1.,  1.,  1.]])



In [17]:

    
5.0 * matrix









    Out[17]:





array([[ 5.,  5.,  5.],
       [ 5.,  5.,  5.],
       [ 5.,  5.,  5.]])

Q. Multiply a matrix by a row of data, without iterating through the matrix rows.



In [18]:

    
np.r_[[1, 2, 3]] * matrix









    Out[18]:





array([[ 1.,  2.,  3.],
       [ 1.,  2.,  3.],
       [ 1.,  2.,  3.]])

Q. Multiply a matrix by a column of data, without iterating through the matrix columns.



In [19]:

    
np.c_[[1, 2, 3]] * matrix









    Out[19]:





array([[ 1.,  1.,  1.],
       [ 2.,  2.,  2.],
       [ 3.,  3.,  3.]])

Numpy allows operations to happen at a lower level without iteration, a process called vectorization.



In [ ]: