Last minute

Question 1

MLE vs Bayes Inference

Answer 1

MLE treats the params as unknown constants and tries to find their ML value.

Bayes uses prior knowledge to estimate parameter values

Question 2

Cross validation

Answer 2

validation set is used to see how your model generalizes before using test set

Question 3

E step of EM algorithm

Answer 3

find the expected value of the hidden variables based on current param values

Question 4

M step of EM algorithm

Answer 4

recompute the ML value of the params based on the value of the hidden variables and observed data

Question 5

Gaussian parameters

Answer 5

k, means covariances and weights

Question 6

p(x) = sumToK p(x | G(i)) * P(Gi)

Answer 6

Gi - components
P(Gi) priors
P(x|Gi) - component densities
p(x) the mixture model

Question 7

Which lines do you remove from adaboost?

Answer 7

Adaboost remove line 7 -12

Question 8

What is p in the adaboost question

Answer 8

p is the probability of sampling each data point

Question 9

What is B in the adaboost question

Answer 9

B controls how much the probabilities are adjusted each round

Question 10

What effects do bags have on ensemble learning?

Answer 10

more bags = high bias, low variance

less bags = low bias, high variance

Question 11

In the 2018 pseudocode what is N?

Answer 11

Number of data points

Question 12

In the 2018 pseudocode what is d?

Answer 12

Number of features

Question 13

What’s a benefit of online in GD?

Answer 13

Converge faster, updates weights one-by-one as inputs are seen.

Question 14

11.1 What is the weight?

Answer 14

wTH

Question 15

What do you need to replace in 11.1?

Answer 15

6,7,8 to yi <= sigmoid(vTiZ)

Question 16

What’s bias and variance?

Answer 16

Bias is the difference between the predicted data points and the actual data points

Variance is how scattered the predicted values are from the actual value