MOL Bubi Challenge 2015
Results Congratulations for the winners!
Final leaderboard
Veress Tamás Task 2 győztes felajánlása 100,000 Ft-ot a
3. feladat további díjaira fordítunk
A MOL Bubi bérletet a mindkét feladaton dobogós
Kerepesi Csaba nyerte
Busiest route prediction [BRP] 1.Bubilyze! éves MOL Bubi bérlet és 100,000Ft Balogi Csilla, Fülöp Árpád, Bíró Márton, Badics Milán, Windhager-Pokol Eszter Balabit, Clementine, I-insight, BME-BCE-ELTE
2.Adatikus éves MOL Bubi bérlet és 75,000 Ft Kerepesi Csaba ELTE IK Doktori Iskola
3.DatBubi 50,000 Ft Magyar Zsanett, Gálai Bence, Horváth Benedek, Stein Dániel BME Mérnök informatikus MSc
Docking station demand prediction [DSDP] 1.Tamas Veress Tamás (Thaiföld) Schneider Electric
2.Brewster Analytics 75,000 Ft Fülöp Bálint, Fülöp Gergő, Vajna Szabolcs BME TTK Fizikai Tudományok Doktori iskola
3.Adatikus Kerepesi Csaba ELTE IK Doktori Iskola
50,000 Ft
Open research task 1
Czeglédi Imre
100,000 Ft
BME VIK, BSc szakdolgozat, konzulens: Nagy István “Adatelemzési megoldások a budapesti közbringarendszerben”
2-3
Bardóczi Alexandra
50,000 Ft
BME Építészmérnöki Kar, TDK, konzulens: Kádár Bálint A városi közbringarendszer működésének elemzése Big Data módszertan alkalmazásával
2-3
Stippinger Marcell
50,000 Ft
BME TTK Útvonal-eloszlások vizualizációja
4-5
Boza István, Molnár Péter, Szabó Tibor 25,000 Ft Degrees of Freedom - ELTE, Társadalomtudományi kar “Network visualization of the Bubi system”
4-5
Magyar Zsanett BME VIK, BSc szakdolgozat, konzulens: Kazi Sándor “Előrejelzés kerékpárkölcsönzési adatok alapján”
25,000 Ft
Participants
Registered teams [>65]
Registered teams - affiliation
Submissions
Submissions
Dataset https://dms.sztaki.hu/bubi
Dataset [2015.01.01 - 2015.05.31]
Travels bicycle_id start_time end_time start_location end_location Station info place_id place_name lat lon num_of_rack datetime_start datetime_end
Weather date and time temperature in C humidity in % wind speed in kph wind direction in degrees wind direction description pressure in mBar visibility in Km wind chill in C fog rain snow hail thunder
Preprocessing
Operator travels / re-allocations are excluded Invalid travels are excluded Weather info from http://www.wunderground.com/weather/api
Training and test
Dataset, codes, further info Homepage https://dms.sztaki.hu/bubi/#/app/home
Dataset will be available for registered participants at https://dms.sztaki.hu/hu/letoltes/mol-bubi-analytics-challenge-training-and-test-data
● Check the test set ● The full dataset is available for further research - connect us if you have a research plan
Challenge codes [e.g. evaluator] @GitHub https://github.com/bubichallenge/challenge-codes
Tasks https://dms.sztaki.hu/bubi
Task 1 - Busiest route prediction Predict daily toplists of the most frequent directed station pairs [routes] Evaluation: daily nDCG@100
nDCG is high, if frequent routes are ranked high in your prediction Margit sziget
Kálvin tér
Nyugati pu.
Kálvin tér
Oktogon
Nyugati pu.
Nyugati pu.
Jászai Mari tér
Task 2 - Docking station demand prediction Predict when will a station become empty Demand = #taken - #docked
Goal: predict the daily maximum demand Evaluation: root mean squared error
Task 3 - Open research
Data Science @ SZTAKI
Data Science @ Sztaki "Big Data - Momentum" research group of the Hungarian Academy of Sciences (MTA SZTAKI) Upcoming data mining challenges RecSys Challenge 2016 with XING ECML PKDD Discovery Challenge 2016 with OTP
Data science course at the Aquincum Institute of Technology
1st prize for DSDP (Veress Tamás) Observed demand/decrease data (https://dl.dropboxusercontent.com/u/59801312/raw_decrease.csv) Weather: average temperature and number of rainy hours Always filled the gaps with the previous best prediction Replaced demand by exponential moving average Robust regression and random forest (rlm and rf in R) Iterating (use the future, overfitted results) 1.
Filling gaps with the best prediction
2.
Regenerate predictors including May
3.
Run predictions for each station
Final submission has several parametric and non-parametric tweaks.
