Testing AI code with your own cases

A shop promises free delivery on orders of £50 or more. An assistant writes the rule, and it looks right. A customer spends exactly £50 and is charged £4.

One test with £50 would have caught it in a second. This lesson is about choosing that test yourself.

A test case is one input and the result you expect. Work out the expected result from the requirement, before you run the code. If you read the answer from the code first, you only test that the code agrees with itself.

In Python, assert checks one case. If the check is false, the program stops and says which one failed.

# Requirement: delivery costs 4 pounds.
# Orders of 50 pounds or more get free delivery.
# The assistant wrote this:
def delivery_charge(total):
    if total > 50:
        return 0
    return 4

# Our own test cases, written from the requirement.
assert delivery_charge(20) == 4, "normal case"
assert delivery_charge(80) == 0, "normal case, free"
assert delivery_charge(50) == 0, "boundary: exactly 50"
print("All cases pass")

# Press Run. Which case fails? Fix the function, then run again.

Good cases come in three kinds. A normal case is a value people use every day. A boundary is the exact value where the rule changes, and the values just either side. An edge case is odd: empty, zero, negative or very large.

Assistants are good at the normal case. Their faults hide at the boundary and the edges.

# Requirement: a password needs 8 or more characters.
def long_enough(password):
    return len(password) >= 8

assert long_enough("sunflower1") == True, "normal, long"
assert long_enough("cat") == False, "normal, short"

# Add three cases of your own under this line:
# one with exactly 8 characters, one with 7, and an empty one.

print("All cases pass")