Continuing our exploration of Data Science from Scratch by Joel Grus (ch2). You'll note that the book emphasizes pure python with minimal libraries, so we may not see much of NumPy in the book. However, since NumPy is so pervasive for data science applications, and since
lists and NumPy
arrays have much overlap, I think it would be useful to use this opportunity to compare and contrast the section, knowing that for most of the book, we'll be using Python lists.
Lists are fundamental to Python so I'm going to spend some time exploring their features. For data science,
NumPy arrays are used frequently, so I thought it'd be good to implement all
list operations covered in this section in
Numpy arrays to tease apart their similarities and differences.
Below are the similarities.
This implies that whatever can be done in python
lists can also be done in numpy
arrays, including: getting the nth element in the list/array with square brackets, slicing the list/array, iterating through the list/array with start, stop, step, using the
in operator to find list/array membership, checking length and unpacking list/arrays.
# setup import numpy as np # create comparables python_list = [1,2,3,4,5,6,7,8,9] numpy_array = np.array([1,2,3,4,5,6,7,8,9]) # bracket operations # get nth element with square bracket python_list # 1 numpy_array # 1 python_list # 9 numpy_array # 9 python_list[-1] # 9 numpy_array[-1] # 9 # square bracket to slice python_list[:3] # [1, 2, 3] numpy_array[:3] # array([1, 2, 3]) python_list[1:5] # [2, 3, 4, 5] numpy_array[1:5] # array([2, 3, 4, 5]) # start, stop, step python_list[1:8:2] # [2, 4, 6, 8] numpy_array[1:8:2] # array([2, 4, 6, 8]) # use in operator to check membership 1 in python_list # true 1 in numpy_array # true 0 in python_list # false 0 in numpy_array # false # finding length len(python_list) # 9 len(numpy_array) # 9 # unpacking x,y = [1,2] # now x is 1, y is 2 w,z = np.array([1,2]) # now w is 1, z is 2
Now, here are the differences.
These tasks can be done in python
lists, but require a different approach for NumPy
array including: modification (extend in list, append for array). Finally, lists can store mixed data types, while NumPy array will convert to string.
# python lists can store mixed data types heterogeneous_list = ['string', 0.1, True] type(heterogeneous_list) # str type(heterogeneous_list) # float type(heterogeneous_list) # bool # numpy arrays cannot store mixed data types # numpy arrays turn all data types into strings homogeneous_numpy_array = np.array(['string', 0.1, True]) # saved with mixed data types type(homogeneous_numpy_array) # numpy.str_ type(homogeneous_numpy_array) # numpy.str_ type(homogeneous_numpy_array) # numpy.str_ # modifying list vs numpy array # lists can use extend to modify list in place python_list.extend([10,12,13]) # [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 12, 13] numpy_array.extend([10,12,13]) # AttributeError: 'numpy.ndarray' # numpy array must use append, instead of extend numpy_array = np.append(numpy_array,[10,12,13]) # python lists can be added with other lists new_python_list = python_list + [14,15] # [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 12, 13, 14, 15] numpy_array + [14,15] # ValueError # numpy array cannot be added (use append instead) # array([ 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 12, 13, 14, 15]) new_numpy_array = np.append(numpy_array, [14,15]) # python lists have the append attribute python_list.append(0) # [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 12, 13, 0] # the append attribute for numpy array is used differently numpy_array = np.append(numpy_array, )
lists and NumPy
arrays have much in common, but there are meaningful differences as well.
Now that we know that there are meaningful differences, what can we attribute these differences to? This explainer from UCF highlights performance differences including:
I'm tempted to go down this 🐇 🕳️ of further
array comparisons, but we'll hold off for now.
For more content on data science, machine learning, R, Python, SQL and more, find me on Twitter.