Roadmap

All activities accounted for in fall 2019 rating.

Plan

  • Calendar and deadlines
  • Assignments
  • Quizzes & Live sessions
  • Kaggle Inclass Competition “Alice”
  • Kaggle Inclass Competition “DOTA 2 winner prediction”

Calendar and deadlines

Google Calendar with all deadlines.

Current deadlines (see also the assignments page):

  • September 15 - A1
  • September 27 - Quiz1
  • October 6 - A2
  • October 25 - Quiz2
  • October 27 - A3 & Alice
  • October 30 - Alice solutions
  • November 15 - Quiz3
  • November 17 - A4
  • November 18 - DotA submissions
  • November 20 - DotA solutions

All deadlines are 20:59 GMT (London time).

Assignments

Assignments are announced in the #mlcourse_ai_news channel in ODS Slack team (pinned items). Also, links to fresh assignments are provided in the Readme file of the course repository and on mlcourse.ai/assignments. Apart from that, you can practice with demo assignments, don’t confuse them with “real” ones. Rough plan for fall 2019 assignments is the following:

  • A1. Pandas and data analysis
  • A2. 2 parts. Decision trees, random forest, and gradient boosting. Beating a baseline in a competition (gradient boosting)
  • A3. 2 parts: Beating baselines in a competition (linear models). Logistic regression vs. Random Forest
  • A4. Time series analysis with Python

If stuck with assignments, check course video lectures.

Quizzes & Live sessions

Quizzes are intended to check your understanding of theory behind the applied approaches. Will be announced in the #mlcourse_ai_news channel as well.

  • Q1. Decision trees, random forest, and gradient boosting. From Sept 9 to Sept 27
  • Q2. Linear models. From Oct 7 to Oct 25
  • Q3. Time series, unsupervised learning. From Oct 28 to Nov 15

Deadline for each quiz is Friday, each quiz is followed by a live session on a Saturday morning. During live session we go through quiz questions and have agile debates.

Kaggle Inclass Competition “Alice”

In this competition, you’ll be solving the task of user identification tracking his/her sequence of visited websites. Lets call it “Alice”, because we’ll be classifying whether a person is some Alice or somebody else. The competition is held together with Yandex and MIPT specialization (so please don’t spoil the competition by sharing high-performing Kaggle Notebooks).

Rules:

  • Deadline is October 27, 2019
  • You can make maximum 5 submissions a day, and the competition is individual (that is, 1 person per team, team mergers are not allowed)
  • In case you want to get credits, you need to rename your team (of 1 person) in full accordance with your name in course rating
  • if you overfitted and dropped several paces down on private LB, no offense, it’s life, only private LB is used to calculate credits for top-10 (this is not the case for A3 - there you have to beat baselines w.r.t. Public LB, it’s easier)
  • Up to October 30, those who managed to beat all benchmarks should upload their reproducible solutions in the .py format (python script) here

Rules

Necessary conditions to get credits:

  • You need to stick to competition rules
  • Beat all Yorko’s baselines on private LB
  • A reproducible solution should be submitted in a specified period
  • Team name exactly corresponding to some entry in the course rating

Credits

  • 1 place – 40 credits
  • 2 place – 30 credits
  • 3 place – 25 credits
  • 4-10 place – 20 credits

Kaggle Inclass Competition “DotA 2 winner prediction”

In this competition you are going to utilize all knowledge and skills acquired during the course session. Discussions are held on Kaggle itself, here is an introductory thread.

Rules

  • Deadline for submissions: 2019 November 18, 20:59 GMT
  • Teams up to 4 people are allowed. You need to name your team like “mlcourse_ai_fall2019_team_<id1>_…<id4>“, where ids are positions of team members in the course rating as of November 18th. For a team of 1 person valid team namings are either this or the participant’s name in the course rating (matching precisely)
  • Till November 20, you have time to upload your team’s reproducible solution here
  • The results of the competition and the final course rating will be published on 2019 November 22.

Credits

Grading is as close as possible to “real” Kaggle competitions, but we don’t penalize for team size, each team member gets the specified amount of credits

  • 1 place – 60 credits
  • 2 place – 50 credits
  • 3 place – 40 credits
  • 4-10 place – 30 credits
  • top 10% - 20 credits
  • top 20% - 10 credits