toxic_comments

Toxic Comment Classification Challenge | Kaggle

https://www.kaggle.com/c/jigsaw-toxic-comment-classification-challenge

Models:

• singel layer CNN

• multi-channel CNN

• DPCNN

• textCNN

• textRCNN

• bi-LSTM

• bi-GRU

(ToDO:XGBoost, NB-SVM, Logistic regression -> refer to public kernels)

Ensembling methods:

• average

• weighted average

• geomean

• rank averaging

• blending

References:

• http://ruder.io/deep-learning-nlp-best-practices/index.html#lstmtricks

• https://zhuanlan.zhihu.com/p/28923961?utm_medium=social&utm_source=wechat_timeline&from=timeline&isappinstalled=0

• https://zhuanlan.zhihu.com/p/25928551

• https://arxiv.org/pdf/1510.03820.pdf

• http://riejohnson.com/paper/dpcnn-acl17.pdf

• https://mlwave.com/kaggle-ensembling-guide/

• https://github.com/neptune-ml/kaggle-toxic-starter/tree/master/best_configs

• https://becominghuman.ai/understanding-and-coding-inception-module-in-keras-eb56e9056b4b

Note:

These are testing codes. There may be incompatibel errors in iPython notebook due to frequent changes on Python files.

---

What the winners share

• No.1: https://www.kaggle.com/c/jigsaw-toxic-comment-classification-challenge/discussion/52557
• No.2: https://www.kaggle.com/c/jigsaw-toxic-comment-classification-challenge/discussion/52612
• No.5: https://www.kaggle.com/c/jigsaw-toxic-comment-classification-challenge/discussion/52630
• No.25: https://www.kaggle.com/c/jigsaw-toxic-comment-classification-challenge/discussion/52647
• No.34: https://www.kaggle.com/c/jigsaw-toxic-comment-classification-challenge/discussion/52645
• single model: https://www.kaggle.com/c/jigsaw-toxic-comment-classification-challenge/discussion/52644