Skip to content

Commit

Permalink
Update README.md
Browse files Browse the repository at this point in the history
  • Loading branch information
jxtrbtk authored Sep 28, 2019
1 parent 683daf8 commit 2e60722
Showing 1 changed file with 2 additions and 2 deletions.
4 changes: 2 additions & 2 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -3,13 +3,13 @@ Work done for [Kaggle](http://www.kaggle.com) competitions.

[>>> Kaggle profile](https://www.kaggle.com/jtbontinck)

## APTOS 2019 Blindness Detection
## APTOS 2019 Blindness Detection [>>>](https://github.com/jxtrbtk/kaggle/tree/master/aptos2019-blindness-detection)
Detect diabetic retinopathy from retina images taken using fundus photography under a variety of imaging condition.
Ingredients : Pre-trained CNNs, Efficient-Net, Ordinal Regression, Pseudo Labelling (semi-supervised learning), Multiple training, Test Time Augmentation, trust your CV !

Public kernels :
- [CNN/XGB - End-to-end (0.11)](https://www.kaggle.com/jtbontinck/cnn-xgb-end-to-end-0-11) : re-train a pretrained DenseNet 161 on competition data (5 folds), extract features and then use XBG for prediction + TTA (x2)
- [CNN/XGB - End-to-end (1.54)](https://www.kaggle.com/jtbontinck/cnn-xgb-end-to-end-1-54) : re-train a pretrained DenseNet 161 on **extrenal+competition** data, fine tune training on competition data only, extract features for XBG for prediction, blend CNN and XGB results + TTA on CNN (x4) - The final score = 0.910, corresponds to 758/798 leaderboard positions. There is an overlap between the first phase training data and the second one, so I did not select this kernel as a final submission as the CV was not correct, unfortunately)
- [CNN/XGB - End-to-end (1.54)](https://www.kaggle.com/jtbontinck/cnn-xgb-end-to-end-1-54) : re-train a pretrained DenseNet 161 on **extrenal+competition** data, fine tune training on competition data only, extract features for XBG for prediction, blend CNN and XGB results + TTA on CNN (x4) - The final score = 0.910, corresponds to 758/798 leaderboard positions. *(There is an overlap between the first phase training data and the second one, so I did not select this kernel as a final submission as the CV was not correct, unfortunately.)*

## LANL-Earthquake-Prediction [>>>](https://github.com/jxtrbtk/kaggle/tree/master/LANL-Earthquake-Prediction)
Forecasting earthquakes for Los Alamos Nationl Laboratory.
Expand Down

0 comments on commit 2e60722

Please sign in to comment.