Applied Probability Models for CS





Lecturer: Prof. Ido Dagan Email:

Class Hours: Thursday 10:00-12:00

Exercise Grader: Vered Shwartz Email: vered1986@gmail.coma




Final exercise grades are available.


Grades for exercise 3 are available below. You should have gotten the feedback for your reports to your email.


Exams from previous years are available here.


Exercise 3 submission deadline is extended to 18.01.18. See comment about the exercise below.


You may use numpy in exercise 3.


Grades for exercise 2 are available below. You should have gotten the automatic check feedback to your email. The final grades are the ones written in the spreadsheet. I had to correct the format of the output files for some students because they failed the automatic grading, and I reduced 10 points for bad format. Students who got 0 in the automatic grading for submitting no output file - please email me the output file (one for each pair of students) until Tuesday, January 2nd, and I will correct your grade. The solution will be published on Wednesday.


Exercise 3 is available (see below).


Exercise 1 solution and grades are available (see below).


Exercise 2 is available (see below).


Exercise 1 is available (see below).


Lecture notes from previous years here are available here.


Welcome! See basic probability equations summary page.



Make sure to submit your exercises on time! Points will be taken off for late submissions.

Programming exercises should be submitted via the Submit web interface. Please make sure that you can login to this system.

Ex1 is to be submitted individually by each student. You are encouraged to do the rest of the exercises in pairs.



Ex3 Dataset Underflow and smoothing in EM Ex3 Grades

Due on 14.01.18 18.01.18. This exercise should be done in pairs.

A note about computing the mean perplexity (question (2) in the report): in the previous exercise you used it to compare between language models, but perplexity is not restricted to language models. Perplexity is a measurement of how well a probabilistic model predicts a sample, and it is used to compare between probabilistic models. In this exercise, you need to use it to measure how well your model predicts articles' topics. Compute it using the model's log likelihood and normalize it by the size of the dataset (the number of words): 2^(-1/N * log-likelihood) (or e if you've used ln for the likelihood computation).


Ex2 Download Ex2 Solution Ex2 Grades

Due on 21.12.17 (submit via Submit). This exercise must be done in pairs.


Ex1 Download Ex1 Solution Ex1 Grades

Due on 30.11.17 in class (submit hard copies only NOT via Submit). This exercise should be done individually by each student.