given two arrays A and B, find top K pairs (a,b), such that a from A and b from B and a+b is among the first K largest sums
Yelp (data mining engineer):. 1point3acres.com/bbs
1.design a procedure to predict click through rate
2.design a procedure to predict review upvotes
3.longest palindrome substring
followup: what if I only have isPalindromeOdd
4.ispalindrome, iterative way to do it
followup: modify the program such that we can skip some chars and see if the remaining is palindrome
2sum, return true or false, there might be duplicates in array
return all the possible combination of indices
Your friend told you that he had at least one girl, what is the probability of he having two girls -google 1point3acres
You then visited your friend, a girl answered the door, what is the probability of he having two girls
in simple linear regression, if you duplicate and double the data, how are beta, t-stats and R^2 gonna change
describe how you are gonna predict future returns based on historical data/ what data? models? procedures?
Linkedin (sde ML track):
1.what is the problem of using KNN when two of the features are highly correlated
3.#In this problem, we have several houses placed on a street.
#We'd like to paint each house either red, green, or blue.
#The amount of money it costs to paint a specific house a specific color varies (maybe the owner already has some old paint that he can use, or parts of his house are already painted that color).
#A house cannot be painted the same color as one of its neighbors
#The goal is to paint all of the houses for the minimum total cost.
#R 2 2 6 4 2
#G 0 5 7 1 1
#B 1 1 2 0 4
# if you had k colors instead of 3
# O(kn). more info on 1point3acres.com
5.讲讲Bayes statistics 的general idea
6.什么是overfitting? how to prevent overfitting? what are common techniques?
7.how to train a logistic regression?
1.Prefix notation expression evaluation
["*",3,2] --> 3*2=6
["+","-","*",3,2,1,4] --> (3*2)-1+4 = 9
What is the least squre estimate for coefficient in linear regression
write update function for SGD
how to choose learning rate lambda
how to prevent overfitting
what numerical issue will we get when training LASSO