Paul's Python CoursePPC

The perceptron

Not met yet

A perceptron multiplies each reading by a weight, adds them up and adds a bias. If the total is above 0 it says 1, otherwise 0. It learns one example at a time: when it is right nothing changes, and when it is wrong every weight moves by rate * error * reading, where error = label - guess. It draws one straight line, so it can learn any rule a straight line can split, and no other.

Try it

main.py

Output

Press Run and what your program prints shows up here.

Easy to get wrong

  • Thinks a neuron starts out knowing which readings should count against a yes, before it has learned anything.
  • Thinks a weighted sum of exactly 0 counts as above the threshold, so the perceptron says 1.
  • Thinks a perceptron's weights change after every example, even the ones it got right.
  • Thinks one perceptron can learn any yes-or-no rule, given enough epochs.
  • Thinks that on a rule it cannot learn, the perceptron settles on the best straight line and gets nearly everything right.

Where it comes up