Simple Pattern Recognition and Decision Making

We use patterns we have seen to make decisions: if–then rules, rules checked in order, and labelling a new item like its nearest example.

Key ideas

A decision rule says what to do when a condition holds, like “if it is cloudy, carry an umbrella”.
When rules are checked in order, the FIRST rule that fits gives the label; the later rules are not used.
A rule learned from labelled examples must fit every example, not just most of them.
AND needs both conditions to hold; OR needs at least one of them.
A nearest-neighbour model gives a new item the label of the closest labelled example.
A prediction from a pattern, like the next bus time, can still be wrong sometimes.

Remember

“weight > 15 kg”	strictly more: 15 kg itself does not count
Distance in this chapter	difference in the first numbers + difference in the second numbers
Example	(3, 8) to (6, 4): 3 + 4 = 7
Bus every 15 minutes	8:00, 8:15, 8:30, 8:45 → next at 9:00
Sorting into groups	classification

Watch out

Checking a later rule after an earlier one has already given the label.
Treating “more than 15” as including 15.
Adding the coordinates instead of their differences when finding the closest example.

Worked example

(a) A parcel sorter uses: Rule 1: if weight > 15 kg, label "Heavy". Rule 2: otherwise, if length > 40 cm, label "Long". Rule 3: otherwise, label "Small". (Check the rules in order.) A parcel weighs 13 kg and is 10 cm long. Which label does it get?
(b) A parcel sorter uses: Rule 1: if weight > 20 kg, label "Heavy". Rule 2: otherwise, if length > 30 cm, label "Long". Rule 3: otherwise, label "Small". (Check the rules in order.) Parcels (weight, length): (2 kg, 50 cm), (31 kg, 35 cm), (20 kg, 45 cm), (21 kg, 15 cm), (32 kg, 45 cm). How many are labelled "Small"?

Working: (a) 13 is not more than 15; 10 is not more than 40, so Rule 3; (b) Long, Heavy, Long, Heavy, Heavy

Answer: (a) Small; (b) 0