Keeper Ines pins a note to the front desk: "If we add a repair bench, will people have to wait?"
Comet waves the note at the crew. "Let's build a model of the bench! I will put in everything: every toolbox, every chair, the color of the walls."
Wren spreads out his log from last Saturday's trial bench, a borrowed table set up for one day. It shows when each person arrived, and nothing more.
"What do you notice?" he asks. "Does the wall color change how many people come?"
Comet pauses. "Probably not."
Nova hovers over the log and projects a grid of small boxes. "Would you like a hint?" she asks. "Keep only the details that change the answer to Keeper Ines's question."
Comet turns to you. "Data analyst, which details stay?"
Can a simple model predict what will happen next?
This week you will build a model of the repair bench line. You will run it with a die, test it and make it better.
A computational model makes predictions about a process. It bases them on selected data and features.
The amount, quality and variety of the data, and the features chosen, affect how good the model is.
Predictions are tested to check a model. A model you never test is only a guess.
Abstraction reduces complexity by focusing on the main idea. It hides details that do not matter for the question at hand.
Keeper Ines asked about waiting, so the model needs to know when people arrive. The wall color can stay out.
Every detail you drop is also an assumption: you are treating it as if it does not matter. Write your assumptions down so you can test them later.
| Detail on Comet's list | Keep or drop? | Why |
|---|---|---|
| When each person arrives | Keep | Arrivals decide how long the line gets |
| How long the bench is open (6 hours) | Keep | It sets how many 10-minute slots to model |
| The color of the walls | Drop | It does not change who arrives |
| The member ID of each person | Drop | The question is how many, not who |
| What each person brings to fix | Drop for now (an assumption) | Some repairs may take longer; test this later |
| Statement | True or false? |
|---|---|
| A model makes predictions based on selected data and features. | ? |
| The features you choose have no effect on how good a model is. | ? |
| Abstraction hides details that do not matter for the question. | ? |
| A model is checked by testing its predictions. | ? |
| A good model includes every detail of the real world. | ? |
Wren's log splits the 6 open hours into 10-minute slots. That is 6 slots an hour, or 36 slots in all.
The crew's first rule: roll one die for each slot. A roll of 1 or 2 means one person arrives. Any other roll means nobody arrives.
Two of the six faces mean an arrival. So the rule expects an arrival in about one slot out of three.
One third of 36 is 12, so the first rule expects about 12 arrivals.
Nice thinking, data analyst. Tomorrow you will turn the rule into a simulation and run it.