Notes from NM PD

[schanzer]

Meta

• normalize all variable names for training data
• Gradually assess "what kind of learning?" across lessons

AI Intro

• Predictive models: classifiers -- need more blank space for writing the confidence
• Make your own quiz: somethign lighter than "how long will you live?"

Data Driven Algorithms

• In "The first spell checker", the instructions for Inserting say the letter is inserted AFTER each character, but it should be BEFORE (for numbers 4 and 5)
• First Spell Checker - butts out of seats
• First Spell Checker - Teachers felt that having to come up with 5 examples for each word operation was really tedious. Maybe shorten to 3? (@flannery-denny I'd love your opinion here) At the very least, this is just itching to be a "groupwork-at-the-wall" activity
• Maybe describe these operations using kids in a line, rather than jumping straight to letters

Programming with Spelling Suggestions

• glossary link is broken
• sorting results should ignore capitalization

Decision Trees

• card sort for animals
• students (and teachers!) NEED to play "I Spy, Classify" with one another for a few rounds
• the dataset online does not match the workbook page (16!!!!)
• too much typing on Computing Decision Trees. Maybe the answer is a new starter file with those trees pre-filled in
• the confusion matrix workbook page is just giant, slow, and painful. We should fill in about 80% of it for them, so they only have to do the "interesting" parts that require actual math
• confusion-matrix needs to convert the prediction to a string before labeling the column
• paper printout of cards

Regression Models

• When teachers/students come up with scatter plots, have them FIRST draw anonymous, unlabeled scatter plots. Then fit a line, THEN come up with a story
• The range of values in each column needs to be on the "exploring the driving simulator" wbpage
• Need explicit guidance about whether to define the training dataset on the left or right
• reflecting wbpage -- ???
• Add a "which kind of ML is this?" page, to distinguish between Classifiers and Regression

Similarity

• use the match-maker analogy
• the supervised-or-unsupervised page went over SO well, especially when I asked teachers to tell me what kind of supervised learning was used. They wanted more, so add two more -- and include similarity models

Natural Language

• tweak essays so that we get silly results with the base set of columns
• "normalizing text" -- all-cols-similarity(computed, "?") zero dot product
• Ethics of 3rd graders -- change the prompt to include "given what you kniow about how this works.."
• Align table names between image-corpus and text-corpus

Generating text

• add clarifying note that completions is only predicting the next ONE word
• encourage students using words that aren't in the corpus
• At the bottom of "Constructing a Language Model", add fields to calculate decimal and percentage probabilities at the bottom

Chatbots

• need a matching page, connecting Pyret functions to UI in Soekia
• Maybe a starter file using the same documents?
• ":meet soekia" pages
• Add a "which kind of ML is this?" page, to distinguish between Classifiers, Regression, Similarity, AND SLMs

[jpolitz]

I was thinking inserting needed to be “before first and after last”, FWIW, but not sure if both matter