1. Install the required packages listed in the dependencies by using "pip install [name]"
2. Execute the data_download.py file
   • Make sure you see the success message at the end
   • Should the file download fail, follow the instruction at the top of the file
3. Execute any file you want!
   • Each file is completely self contained, since these files used to be notebooks on kaggle.com
   • You can find a description below

• data_analysis.py Gives an overview over the data
• embedding_CNN_LSTM.py Contains a CNN and a LSTM classifier
• meanembedding_SVM Contains a SVM classifier with mean embedding vectorization
• metaembedding_Capsule Contains a Capsule classifier with a meta embedding (External)
• metaembedding_Ensemble Contains a Ensemble classifier with a meta embedding (External)
• preprocessing_comparison Contains a comparison between using and not using preprocessing before the embedding
• tfidf_SVM Contains a SVM classifier with TF-IDF vectorization

• Have around 22 GB of free disk space
• Have 16GB+ of RAM for loading the embeddings
• Have all specified packages & data files
• Recommended: Have a GPU
• Recommended: Have tensorflow-gpu installed

Please make sure to have these python packages installed before runnging the code

pandas
tqdm
textblob
timeit
sklearn
keras
tensorflow
numpy

• All right reserved
• Certain code snippets are taken from different owners. I tried to include all sources.
• This GitHub Repo https://github.com/MarvinThiele/KaggleQuoraChallenge

1. I receive GPU OOM errors when trying to execute the code
   • To fix this issue you can reduce the batch size in the code (at train_pred())
   • If this doesnt fix the issue, it's most likely the embedding layer, which cannot be reduced in size.
   • Use a server with more VRAM in this case
2. My computer freezes and doesn't work anymore
   • This can happen if you are running less than 16GB of RAM
   • This is caused by loading the word embeddings into the memory
3. I have problems not listed here
   • Contact me at [email protected]

• I have taken code snippets & inspiration from other public kaggle notebooks
• Special Thanks to:
  • https://www.kaggle.com/christofhenkel/how-to-preprocessing-when-using-embeddings
  • https://www.kaggle.com/shujian/different-embeddings-with-attention-fork-fork
  • https://www.kaggle.com/sudalairajkumar/a-look-at-different-embeddings
  • https://www.kaggle.com/gmhost/gru-capsule
  • https://www.kaggle.com/vanshjatana/magic-numbers-is-all-you-need-0-692-lb-986394