Machine Learning and Inductive Inference

2021-2022:

26 januari:

small questions:

1. connect 6 concepts with the right scentence about the concept or their definition

2. same as previous question

3. 3 questions where you have to anwser what the concept is.

4. Which of the three has the largest entropy? {a,a,a,a,a,a} OR {a,b} OR {a,a,a,a,a,b}

Big questions:

1. 12 examples given where 4 attributes(2 or three possible values per attribute) predict 1 class (3 possible values). A new instance their 4 attributes are given.

• Classify the new instance according to naive bayes (m=2, q=1/2)

• Classify the new instance according 4NN. (calculate the distance from the examples).

2. The hypothesis "Tony likes pizza" is formulated with 0, 1 or 2 literals at most. The possible attributes that predict the hypothesis are {mushrooms, ham, tomatoes, onion}. Example that is valid"TOny likes pizza with mushrooms and tomatoes" or for 0 literals "Tony likes all pizza". Invalid is"TOny likes pizza with mushrooms or ham".

• show that the VC dimension of this hypothesis space is at most 3.

• show that the VC dimension of this hypothesis space is at least 3.

3. Given 2 literals (something like: p(X,X) <-- q(X,Y)... and p(X,Y) <-- ... (both had a total of 3) )

• how many variables after combining them to claculate the LGG?

• What is the LGG (multiple choice a-g)

4. Calculate the ROC based on the DT below. The proportions in the leaves discribe how the training data is labeled. Assume that a new instance will get classified as positive if the proportion of positives in the leaf is higher then a threshold C. Let the threshold C vary from 0-1 and draw the ROC for this descision tree classifier.