First TPS:

ML questions:

• Pros & cons of cross-validation
• Ways to validate a model other than cross-validation

Probab & Coding:

• Expected number of trials to see all n numbers in an n-sided dice.
• Also code a simulation for a large number of trials with n-sided dice and match it with probabilistic calculations from previous deduction of expected number of trials to see all n numbers in an n-sided dice

Stats question:

• Normal confidence interval around mean question using CLT, I forgot the exact question.

Second TPS:

ML: Given a Random Forest Model that uses FICO score to predict whether or not you should lend. But so far the model only takes into account people with FICO score >= 640 and has never been trained on data for people with FICO score <640. Assume you have all the credit data you need and FICO scores but you can't train on new data of people with FICO score < 640. Will your new model generalize directly? If yes, why and how? If no, why not and what can you do to improve it?

Coding:

Given an N element array, write a function that find all triples (a, b, c), that satisfy the following famous geometric equality: a^2 + b^2 = c^2. assume all elements are unique, positive integers

myArray = [2, 11, 13, 15, 12, 17, 3, 5, 7, 9, 4]
Output: (3, 4, 5), (., ., .), ...