AutoML bootcamp @ ICML July 11, 2015

JavaScript isn't enabled in your browser, so this file can't be opened. Enable and reload.

First name *

Last name *

Email *

Codalab username *

Country *

Country where you currently live

Preferences -- the hackathon track *

Normal: Learn to make submissions with Python

GPU: Participate in the GPU/deep learning group (more advanced track -- good knowledge of Python required)

I will not attend the hackathon and will not attend the free lunch after the workshop -- thus will only attend the workshop/hackathon opening/will be a spectator, so I do not need to do the homework below.

I will not attend at all -- I want to unregister

Python knowledge (we will allocate people into groups with similar experience) *

Standard (basic knowledge of Python)

Advanced -- experience in programming (e.g. I work as software developer/ finish-(ed/ing) CS studies or equivalent, etc), good knowledge of Python.

Beginner -- I never programmed in Python (I will learn at least the Python syntax by myself before the event)

Starting filling the form -- checklist (you need to agree to all) *

1. I will measure (roughly == will be able to provide some indication) the overall time I spent for completing the homework and reading this form.

2. I will complete/confirm all the checkboxes related to this homework, before submitting the form

3. I know the deadlines related to registration given on the hackathon webpage.

4. A computer to do the homework should have at least 8GB of RAM memory. It should still be possible to run it if there are 4GB of RAM (even 64bit OS) using SWAP.

5. If my laptop has less than 6GB of memory, I will: a) restart the computer once I prepare code/datasets to run b) do not open any programs (especially a browser) while executing the code, to have more free memory, c) I will also close all not important programs which started on startup -- open a Task manager and sort programs by memory usage. Close all programs like Skype, GoogleDrive/Dropbox etc.

Required

"Free lunch" compulsory check list (required homework) *

In machine learning there is NO free lunch. To win your free food complete ALL the following prerequisites. This will also guarantee you WiFi access and GPU access, and participation in the hackthon. Numbers in ( ) refer to the comments below.

1. Register to Codalab at http://codalab.org.

2. Join the group https://groups.google.com/d/forum/automl.

3. Download all the datasets of round 1 and 2 from http://codalab.org/AutoML (1).

4. Put all the data zip files [dataname].zip in a directory [datadir] and unzip them. The directory tree should look like [datadir]/[dataname]/[datafiles].

5. Download and install Anaconda for Python 2.7 http://continuum.io/downloads (2).

6. Download sample code from https://sites.google.com/a/chalearn.org/automl/hackathon-icml/sample_code_hackathon2_7.zip?attredirects=0&d=1

7. Unzip the archive in a directory [workdir]/sample_code_hackathon/.

8. Edit run.py such that default_input_dir = [datadir] (3). DO NOT PUT DATA INSIDE THE FOLDER WITH CODE (sample_code_hackathon/), so the datasets are not included in the submission.

9. Run the code run.py from the directory sample_code_hackathon (4). NOTE: this should run about 1/4 of the total time budget, the total time budget is 200 minutes = 10 datasets * 20 minutes. All 10 datasets = round 1 and round 2 datasets. If you have a little of RAM memory (e.g. laptop with 2 GB of free memory), it might use SWAP memory and run too long. If my laptop has less than 6GB of memory, I will restart it now and do not open a browser/programs (as described above) while the code is running.

10. Go in the directory [workdir] (one level up); retrieve the zip file generated automatically (5).

11. Ensure that you ran successfully - submission (.zip) should have predictions for each of the 5 datasets in the /res folder. Of course the .zip should not contain the datasets!

12. Submit the zip file to http://codalab.org/AutoML (6) (we will check this)

13. Ensure that execution time (it is displayed on the leaderboard) of your submission is low (a few seconds, usually should be around 1..4 seconds + a few seconds for possible delays (if you have sample code version 2.6 or earlier, due to zipping it will execute extra 300 seconds, this is fixed in 2.7). If the duration is above 600, it means you run training(s) on the platform -- check again if the predictions for each dataset are in the /res folder

14. If the execution time is still large (above 600.0) and there are predictions in the /res folder, check that the datasets are correct -- these should be datasets from Round 2 (at least, predictions for Round 1 can also be there). Datasets names of round 2 are e.g.:"albert", "dilbert".

15. I confirm that the duration given in 'Duration' column on the leaderboard of my submission is below 600.0 -- else I will fix it now, BEFORE submitting this form (sorry that it is the same as above... -- this is important for the submission to be valid)

Required

Comments

(1) Go to the "Participate" tab, accept the rules, go to "Get data" in the left side menu, scroll down. Complete the pre-requisites by downloading the datasets of Round 1 and 2. It is OK to put them all in the same directory. [datasets round1 are needed for the hackathon, round2 for this homework - good to get both now]
(2) If you already have Anaconda, update such that you have those versions or more recent ones: scikit-learn 0.15.2, numpy 1.9.1, scipi 0.14.
To update Anaconda, type at the prompt:
conda update conda
conda update anaconda
conda update scikit-learn
conda update numpy
conda update scipy
(3) You can also use any text editor. The Spyder editor coming with Anaconda can be started with the Anaconda launcher.
(4) If you use the Spyder editor, just press the green arrow. This should create a subdirectory res/ with the results. The program run.py takes input_dir and output_dir as arguments, but if you have made the change of step 8., you don't need any arguments, you can just run "python run.py" at the shell prompt, else: "python run.py /path_to_data res"
(5) Zip should Include metadata, run.py, lib/ and res/, but not the data.
(6) Go to the "Participate" tab, then to "Submit/view results" in the left side menu.

Time taken --- ONE NUMBER in MINUTES *

Time taken to do all the tasks above (homework and solving any related problems appearing). Include time of code execution (but only one if you did re-running of code for some reasons)

Registration checklist (confirm all) *

0. I confirm the the time above is given in minutes, not hours or other unit

1. If I chose the GPU track: Complete the reading prerequisites http://automl.chalearn.org/hackathon-icml/gpu-track-icml Chapter 2, point 1.

2. I am aware that the Registration on-site will start at 19:00 and I cannot come at a random time during the reception -- 19:00 is the time to arrive. Plan to come earlier if you are not on-site during the day. The number of participants is limited -- people on the waiting list can use my spot (this will be done during the cocktail reception) if I will not appear by 19:00 on-site.

3. I will not ask any questions regarding my problems with the homework on the last day (5th July). If I cannot solve my problems using the Internet, I can ask important questions till 4th July (email above the form on the hackathon webpage)

Required

Comments regarding the instructions (or the form in general)

If you would like to help to improve these instructions, tell us how. Rembember to describe precisely, mention the operating system version, etc. Also tell us if you think we need to describe some step that might not be obvious.

Final standings *

(OPTIONAL) Please add me to the NVIDIA mailing list for latest GPU news

In case it will happen that I will not be able to attend the hackathon, I will inform the organisers about it (address given at the technical questions section on the webpage).

Since the list of usernames, who will be registered to the hackathon, will be available on the hackathon webpage (after we process the registrations), I will be able to ensure whether my registration is definitely valid.

I will bring a fully charged laptop to the hackathon event (on which I will work and on which I already tested that the sample code has run there without problems)

Required

Submit

Clear form

Never submit passwords through Google Forms.

This content is neither created nor endorsed by Google. - Terms of Service - Privacy Policy

Does this form look suspicious? Report

Forms