- NumPy arrays come with many useful methods
- All arithmetic operations that are used on arrays are performed element-wise
- NumPy code is almost always faster than native Python (.append is a notable exception)
NumPy arrays are so useful because they allow us to do math on them very efficiently. For example, NumPy arrays come with many useful methods. One such method is the sum method, which calculates the sum of all values in the array
import numpy as np my_array = np.array([4, 3, 1]) my_array.sum() 8
There are many other methods like this and they are extremely useful. Here is a list of the most commonly used methods.
my_array = np.array([4, 3, 1]) my_array.sum() # Calculate the sum array values 8 my_array.mean() # Calculate the mean of array values 2.6666666666666665 my_array.std() # Calculate the standard deviation of array values 1.247219128924647 my_array.max() # Find the maximum value 4 my_array.min() # Find the minimum value 1
To learn about all array methods you can call the dir() function on any array, which will list all its methods. Alternatively you can check out the documentation for the array https://docs.scipy.org/doc/numpy/reference/generated/numpy.ndarray.html
Another useful property of arrays is that they do math when they appear together with any of the arithmetic operators (+, -, *, /, **, //, %).
my_array = np.array([4, 3, 1]) my_array_plus = my_array + 2 my_array_plus array([6, 5, 3])
Here, the array appeared together with a scalar value, the single number 2. That number was added to each value. However, we can do the same thing with two arrays, if the have the same shape.
array_one = np.array([4, 3, 1]) array_two = np.array([1, 2, 4]) array_plus_array = array_one + array_two array_plus_array array([5, 5, 5])
In this case, addition is again performed element-wise. Each element in array_one is added to a corresponding element in array_two. The fact that the array performs useful math in this context might seem unremarkable but remember how the native Python list behaves.
list_one = [4, 3, 1] list_two = [1, 2, 4] list_plus_list = list_one + list_two list_plus_list [3, 2, 1, 1, 2, 4] array_plus_array = np.array(list_one) + np.array(list_two) array_plus_array array([5, 5, 5])
If you are in full numerical computation mode this behavior of list might seem stupid to you. But remember: Python is a general purpose programming language and list is a general purpose container to store a sequence of objects. There could be anything in those lists and addition might not be a meaningful operation for those objects. This behavior always works, a list can be concatenated to another list regardless of the objects they store. That’s why we have NumPy. Python has to implement objects in a way that suits its general purpose. NumPy implements behavior in a way that we would expect while we do numerical stuff.
A word on performance
This is one of the rare occasions where it is worthwhile to talk about performance. When you are getting started, I strongly recommend against thinking too much about performance. Write functioning code first, then worry about readability, maintainability, reproducibility etc. etc. and worry about performance last (trust me on this one). But some of you will be working with large amounts of data and you will be delighted to hear that NumPy is much faster than native Python.
my_array = np.random.rand(100000) # A large array with 100000 elements my_list = list(my_array) timeit sum(my_list) 18.1 ms ± 801 µs per loop (mean ± std. dev. of 7 runs, 100 loops each) timeit my_array.sum() 90.3 µs ± 6.86 µs per loop (mean ± std. dev. of 7 runs, 10000 loops each)
The native Python version of sum is orders of magnitude slower than the NumPy version. You might have noticed that I created a very large array to demonstrate this. Actually the performance difference will increase with increasing array size, you can verify this for yourself. The take home message here is that whenever you can replace native Python with NumPy, you gain performance. But don’t worry about optimizing your NumPy code. One exception is the .append method, but more on that later.
We learned two essential things and one kind of interesting side-note. The first essential lesson is that arrays come with many methods that allow us to do useful math. We learned some of those methods and as you keep working with NumPy those will become second nature. The second thing we learned is that arithmetic operators are applied element-wise to arrays. This means that a scalar value is applied to each element in an array and whenever two arrays of the same shape appear together with an operator each element is applied to each corresponding element. We will learn the details of array shapes in the next blog post. Finally, we also learned that NumPy code is almost always much faster than native Python code. This is good to know. However, especially in the beginning you should focus on anything but performance.