Kaggle Paris Meetup
Meetup #15:
- Introduction and welcome by Algolia
- Survey and trends in DS for the past year
- Porto Seguro competition (ended 30th nov.)
- Statoil competition (Image Classification)
-
prevision.io Saas ML
2017 summary and trends
- A lot of Meetups organized in the town : Kaggle , Paris ML, NLP, RecSys, Deep Learning, Big Data and IA, Paris Business, ...
- Theory : main conf : ICML Sydney, ICLR (Toulon), NIPS this week in LA
- Next year : ICLR Vancouver, ICML in Sweden - Stockholm, ICLR NIPS in Montreal
- All papers, posters, video (not always) available on the web
- Auto ML / Tooling in the 70% data eng/ 20% coding / 10% DS is autoML only improving the 10% and 20% part :-) ?!
- Studios : a lot of offer from DSS to DIGITS
- Cloud : from mamouth AWS , GCP ML platforms to startup : Predicsys , Prevision.io
"Predict if a driver will file an insurance claim next year"
- Binary Classifier
- Size Train 590k lines, Test 890k - 300 Mo
- Metric : Gini , "gini = 2 * auc -1"
- EDA
- A lot of ensembling, not so much stacking
- Non sense overfitting with blending reusing best single model solutions publicely available
- 5 submissions per day
- $ 25,000 · 5.200+ teams ·
- Ended 30th nov.
- Image classification - Iceberg classifier
- Size : train 1600 imgs , test 8400 imgs - 1,6Go
- Metric : logloss
- EDA , bis
- 2 submissions per day !
- Repo with good accuracy and well documented :
- https://github.com/QuantScientist/Deep-Learning-Boot-Camp/blob/master/Kaggle-PyTorch/iceberg/statoil-iceberg-classifier-challenge-cnn-ver1.py
- 2 months to go
KaggleParisMeetup-15
By bruno16
KaggleParisMeetup-15
Slides for Kaggle Paris Meetup
- 1,439