Functions and imports

def begins a function definition. Its name is count_label. The names inside parentheses are parameters: places for values a caller will supply. Defining the function stores the procedure; it does not yet run its indented instructions. The display after the definition is outside the body. The counter begins anew whenever the procedure is called, so separate calls do not accidentally share its running total.

Run and notice that defining the function does not print a count.

In the call, examples and 2 are arguments: concrete values supplied to the parameters labels and wanted. During this call, labels refers to your six-element list and wanted refers to two. return count ends the call and sends its result back, allowing the assignment to twos to store it. The return is after the loop but still inside the function. Returning inside the loop would stop after the first element.

Trace which values reach each parameter, then run.

Three calls use one definition. Each call creates its own local count and completes before print receives the result. Nesting a function call inside print is like putting the result of one block into another block's input slot. The last call returns zero because zero occurs no times; absence of matches is still a meaningful numerical answer. It is different from returning None, which usually means no value was returned.

Predict all three counts and then run.

The first function shows eight but does not explicitly return a value, so its caller receives None. The second returns eight without displaying it; the later print reveals the stored result. Printing is an output action. Returning transfers a value to the caller, where it can be used in another calculation. A model-preparation function must return prepared data, not merely show it, because the model needs those values as inputs.

Run and distinguish the line produced inside a function from the two lines produced afterward.

Parameters do not guarantee sensible values. This function checks its assumptions before dividing: total must be positive, and correct must be a possible count. or catches either invalid end of the range. A helpful exception explains the input contract rather than quietly returning nonsense. For a small known example, you can calculate the answer yourself and check whether the procedure matches your intention.

Run the valid example, then try total equal to zero and read the deliberate error. Restore valid counts.

A library is a collection of functions other people have written for you to use. A module is one file of such functions; math is a module that comes with Python. from math import sqrt makes its square-root function available under the name sqrt. After that, calling it is exactly like calling your own count_label: a name, then arguments in brackets. Importing only makes code available; it does not download anything new. The libraries this course needs are already set up in the browser, though they may take a moment to load the first time.

Run and compare the argument 9 with the returned result.

NumPy is the library most Python programs use for working with lots of numbers at once. import numpy as np imports it under the short name np, which nearly every example you will ever see uses. np.array([0, 8, 16]) makes a NumPy array: like a list of numbers, but built for maths. The big difference: dividing an array by 16 divides every number in it, in one step, while dividing an ordinary list by 16 is an error. The dot means "belonging to": np.array is the array tool belonging to NumPy, and scaled.shape is the shape belonging to scaled. (3,) means three values in a single row; the comma is how Python writes a group containing one number.

Run and compare every original value with its scaled partner.

scikit-learn, imported under the name sklearn, is a machine learning library, and it comes with a small collection of handwritten digit images. load_digits() returns that collection as one object, here named digits, which holds several parts. The dot picks a part: digits.data is a table with one row per image and 64 brightness values per row, and digits.target holds the correct digit for each row, in the same order. The name X is used for the table of inputs and y for the answers, a habit shared by most machine learning code. .astype("float64") stores the numbers as decimals so that dividing works cleanly, and dividing by 16 turns the original brightness range of 0 to 16 into 0 to 1. Loading data does not train anything.

Run and identify which object is the returned bundle and which names refer to its numerical data.

Your counting function can iterate over this one-dimensional NumPy array just as it iterated over a list. A reusable procedure is useful when its input contract matches another source of data. It does not need to know how the dataset was loaded. The caller supplies the labels and the wanted value; the procedure returns the count. This separation makes errors easier to locate: loading, counting, and displaying each have a clear job.

Run for a chosen digit, then compare a second digit using another call.