ОУФ Машинное Обучение в Питоне

О курсе

Курс читается в 1-2 модулях.

Instructor: Oleg Melnikov

Ассистенты: see Canvas LMS

This course introduces the students to the elements of machine learning, including supervised and unsupervised methods such as linear and logistic regressions, splines, decision trees, support vector machines, bootstrapping, random forests, boosting, regularized methods and much more. The two modules (Sept-Dec, 2020) use Python programming language and popular packages to investigate and visualize datasets and develop machine learning models.

Пререквизиты курса: at least one semester of calculus on a real line, vector calculus, linear algebra, probability and statistics, computer programming in high level language such as Python or R.

Технические требования: Laptop, Internet connection, Chrome web browser, Google Drive, Google Colab.

План курса

Лекции

1. Math Essentials. Intro to Python in Google Colab

2. Ch2. Intro to Statistical learning

3. Ch3. Linear Regression

4. Ch3. k-Nearest Neighbors

5. Ch4. Classification: logistic regression

6. Ch4. Classification: LDA, QDA, KNN

7. Ch5. Resampling methods. CV, Bootstrap

8. Ch6. Linear model selection & regularization

9. Ch7. Non-linear regression

10. Ch7. Non-linear regression-2

11. Ch8. Decision Trees

12. Ch8. Bagging, Random Forest, Boosting

13. Ch9. Support Vector Machines/Classifiers

14. Ch10. Clustering methods. PCA, k-Means, HC

15. Special Topics: tSNE, UMAP, Neural Networks

Итоговая оценка за курс

ОУФ Итоговая оценка = 0.35*HW + 0.1*Q+ 0.05*P + 0.5*(E1 + E2)

HW, exams, and project grades are on the scale of 0-100. Course grade is scaled to 0-10, which is the range used at HSE. There are no blocking grading components.

Assignment submission: All submissions will be done as PDF and IPYNB file via Canvas LMS. Graders will leave feedback in your PDF and execute your IPYNB to reproduce the results.

HW: weekly graded homework (HW) assignments, which will include analysis of datasets, analytical and conceptual problems, and programming assignments. These are to be completed individually.

Exams (E): There will be exams at the end of each of the 4 modules. The examination locations are TBD. An in-class exam is closed book, notes, calculators and phones. Take-home exam is an open book/internet, but no collaboration. Exam questions are different from homework questions: HW deepens your understanding, but the exams measure it. Each exam is cumulative. Do not book travel that conflicts with this date.

Automatic grading policy for Exam 2: If grade to the date of exam 2 (G2E1) ≥ 95% and exam 1 grade ≥ 95%, then G2E1 is used as the grade for exam 2.

Coursework Project (CP) in R programming language is for DSBA/ICEF students only and is administered by LSE/UoC. It is released about 1 November and due about 1 April. Although students are given a 4-5 months window, this exercise is meant to be completed in a few days. Typically, students work on it in Feb/Mar. Details TBD.

In-Canvas Quizzes (Q) are based on lectures, slides, and textbooks. Answers can only be submitted once and cannot be seen thereafter, so please check them carefully before submitting. Questions are shuffled and sampled for each student. So, students will likely see different questions.

Participation (P): this includes your active participation in the course, answering questions of your peers in the Piazza forum, and your attendance of seminars and lectures. Redundant and uninformative posts (for the sake of traffic) may lower participation grade. Please leave meaningful questions and comments. All participation is tracked by Zoom software and Piazza. Attendance of the seminar sessions is required and is graded as participation.

Re-grading: We aim to grade fairly, accurately, and timely. If you believe we made a crude grading error, please notify your TA/GA privately via forum ASAP (within 1 week of the grade’s release). To discourage frivolous appeals, we reserve the right to deduct a 2-5% of the grade, if your appeal lacks a strong justification or the benefit fails to exceed 2-5%. Be sure it is worth the mutual effort.

Make up policy: If you miss an exam with a valid/verifiable excuse (be prepared to demonstrate), contact instructors ASAP to reschedule the exam. Please mind that making exceptions is difficult and time consuming and can only be done before exams/solutions are distributed. Typically, a verifiable medical emergency is a valid reason, but travel and conferences are not. Other assignments cannot be made up. It is the student's responsibility to start their work early, so as to hedge against any unforeseeable life event.