alexeygrigorev / Nips Ad Placement Challenge
The winning solution to the Ad Placement Challenge (NIPS'17 Causal Inference and Machine Learning Workshop)
Stars: ✭ 33
Labels
Ad Placement Challenge
The winning solution to the Ad Placement Challenge by Criteo for the NIPS'17 Causal Inference and Machine Learning Workshop
- Report: https://arxiv.org/abs/1712.01913 (also in
report/) - Slides: https://www.slideshare.net/AlexeyGrigorev/ad-placement-challenge
- Video: https://www.youtube.com/watch?v=d5I0baV51vk
The solution is simple:
- Take the data as is, train a FTRL model (using libftrl-python)
- Post-process the predictions:
- Apply the sigmoid fuction to the predictions and scale the result by some constant value
- For each group add +15 to the max value
Environment:
- ubuntu 14.04
- 8 cores, 32 gb of ram
- anaconda with python 3.5.2
- additional packages not from anaconda:
tqdmandftrl
Running the solution
mkdir tmp-
python 01-cv-split.py# takes ~22 mins -
python 02-prepare-data.py# ~17 mins -
python 03-train-model.py# ~25 mins -
python 04-predict.py# ~4.5 hours gzip pred_ftrl.txt- submit
pred_ftrl.txt.gz(~1.5 hours)
Note that the project description data, including the texts, logos, images, and/or trademarks,
for each open source project belongs to its rightful owner.
If you wish to add or remove any projects, please contact us at [email protected].