All labs

Chance machines

Count a weather diary into a table of chances, forecast two days ahead, find the long-run balance, then build the simplest possible language model from three sentences.

Briefing

You start with a three-week weather diary and end with a tiny language model. In between you turn counts into chances, forecast a few days ahead, find the balance a long run settles into, and let a table of word-pairs write a new sentence.

You'll learn to

  • Turn counts into a table of next-step chances
  • Forecast several steps ahead by multiplying through the table
  • Find the long-run balance a chain settles into
  • Build a word-level model and see why it's the ancestor of today's
0 of 6 steps done0%
Start: Count the days
Before the chance machineOptional warm-upChances from counting, why they add up to 100%, and when to multiply and when to add. Ten minutes.

  1. Count the days

    Up next

    A table of what comes next

  2. Two days ahead

    Multiplying by the table

  3. The long run

    Where the chances settle

  4. Count the words

    The same table, made of words

  5. A new sentence

    Let it talk

  6. A very small language model

    From Markov to chatbots