GitHub - Kirupakaran/Toxic-comments-classification: Udacity mlnd capstone

Branches Tags

Name		Name	Last commit message	Last commit date
Latest commit History 14 Commits
.gitignore		.gitignore
LICENSE		LICENSE
Machine Learning Engineer Nanodegree - Final report.pdf		Machine Learning Engineer Nanodegree - Final report.pdf
NBSVM by Jeremy Howard.ipynb		NBSVM by Jeremy Howard.ipynb
Readme		Readme
Toxic comments - benchmark.ipynb		Toxic comments - benchmark.ipynb
Toxic comments - lstm - fasttext.ipynb		Toxic comments - lstm - fasttext.ipynb
Toxic comments - lstm - glove.ipynb		Toxic comments - lstm - glove.ipynb
Toxic comments - starter-with preprocessing.ipynb		Toxic comments - starter-with preprocessing.ipynb
Toxic comments - starter.ipynb		Toxic comments - starter.ipynb
launch.sh		launch.sh
lstm_starter_model.h5		lstm_starter_model.h5
model.h5		model.h5
proposal.pdf		proposal.pdf

Repository files navigation

Readme

Toxic comments classification - Udacity nanodegree final project

Requirements:
python 3
scikit-learn
keras 
seaborn (for visualisation)

Used Nvidia GTX 1060 and cuda 9 to build the model

Used https://github.com/ufoym/deepo as the base docker image and modified it to add seaborn


Data can be obtained from:
https://www.kaggle.com/c/jigsaw-toxic-comment-classification-challenge/data - training and test data
https://github.com/t-davidson/hate-speech-and-offensive-language - twitter data used for validation
https://s3-us-west-1.amazonaws.com/fasttext-vectors/crawl-300d-2M.vec.zip - fasttext embedding