Regression Rewrite

[flannery-denny]

• Rework driving game so that it doesn't rely on claude input and is more responsive to input
• Decide whether to keep animals data or rewrite whole thing to focus on cars
• We want to know how to turn the steering wheel? Obvious 2D relationship between sharpness of curve and degrees of the steering wheel. But there are other variables. Other factors that matter:
  -- Which lane you’re in (offset)
  -- Heading error - the road is going this way, but you’re going that way - how far off the angle of the road is your car facing
  -- Speed
• Confirm we are accomplishing listed learning goals:
• Explain that supervision is the process of humans providing information on known inputs that the machine learning algorithm can use to extrapolate and make predictions for unknown inputs.
• Describe the function that automated the steering of the very first self-driving car.

Regression lesson goals

• Line of best fit expresses a relationship between two variables (old)
• Finding the Line of best fit means training a model that allows us to make predictions (new)

Regression 1 todos:

• Save Desmos activities for multiple regression lesson
• Use lr-plot and not lr-coeffs or simple-s. Decide whether it makes sense to replace lr-code with lr-model
• Either eliminate references to Kujo or scaffold them better.
• Start with an open end question
• Which variables might have an impact on adoption time?
• use lr-plot for age-v-weeks and pounds-v-weeks
• Questions about interpreting
• Lr-model can be plugged into fit-model and show that it matches lr-plot?
• End lesson1 with
• You just trained a regression model to predict how long an animal that is 5yo will stay at the shelter. Is it right all the time? Is it close all the time?
  -- Predicts that an animal of __ age takes __ to get adopted. It could be that Felix is just so old that it’s throwing off the line, or Kujo is so heavy that it’s throwing off the line.
  -- What about a point on the line that doesn’t exist? That’s a hallucination in AI. In our case the output is just a number, in other contexts it might be a whole sentence or Supreme Court case or advice that doesn’t exist. Still not real. High school kids are handing in book reports with citations that don’t exist. Lawyers have gotten screwed because they are citing super court cases in front of a judge that never existed. A guy recently died because an AI told him that missing Zanex and alcohol is fine...
• linear regression is great if there’s only one variable, but life is usually more complicated.
• What if age and pounds both contributed to adoption time? What would that even look like?
• We want something like lr-plot that works with multiple variables.

Multiple Regression lesson goals

• Each data point can exist in an unlimited number of dimensions.
• Each scatter plot is a shadow of the data and looking at it from another dimension might give us a very different scatter plot
• Regression algorithm works in multiple dimensions, just can’t see it, and pyret doesn’t know how to make a graph in more than 2D.
• We can model much more complicated relationships because regression works in multiple dimensions

Mutliple regression todos:

• launch with cylinder projection
• show screenshot of 3D scatterplot and description of 2D scatterplots as shadows of the data that exists in multiple dimensions before working with the 3 Desmos files
• confusing things to clarify
• Add something about subscript notation of slope and y-intercept
• thinking behind function names - currently the starterfile comments are in the opposite order that they usually are - we should surface "point of view" before train-pov
• train-pov(1) produces a different table every time, so we can save d=train-pov(1) and then save x=train-pov(1) and they will be different.
• what do the numbers in the table mean?
  --Left negative / Right positive
  --Angle measures are relative to the direction the vehicle is facing
• animal names are in the ID column
• Connect questions about which plane we're looking at to directions for labeling this diagram