1. Amrit Prasad Phuyal
2. Dipendra Raj Panta

• Windows 11 pro
• Nvidia gtx 1050 Ti
• Core i7 ,16 gb Ram
• Python 3.9.7 (64bit)
• Anaconda Python (for packages) (latest)
• Vscode with Jupyter extension (latest)

• Create virtual environment

• Activate virtual environment

• Install dependencies ( use requirement.txt file)

• For GPU usage

  • (Local only) Install Tensorflow , Tensorflow-gpu , Keras and Keras-gpu from Anaconda repository that will install necessary Library (cuDNN) and Nvidia Kernel (Cuda).
  • (Google Colab) no need to install these libraries but don't forget to select GPU as runtime.

• Data preparation (cleaning and stemming)

  • (Local) All Data has to be in "data" folder in your root directory.
  • (Google Colab) Data has to be in "data" folder in your root directory.and change the path accordingly in notebook.
  • Run prepare_data.py file or prepare_data.ipynb Notebook. It takes time to finish.
  • New files train_clean.csv and test_clean.csv will be created in "data" folder.

• Training

  • We will use train_clean.csv from data folder to train model
    • Run train.py file or train.ipynb Notebook. It takes time to finish.If your GPU supports increase the batch size for faster training.
  • In the process following files will be generated in root folder
    • train_clean_counts_word.csv >> To view no of words in each abstract and help to determine suitable value for MAX_SEQUENCE_LENGTH
    • word_index.csv >> To view index number assigned to each word
    • word_counts.csv >> To view the number of times a word is repeated and determine the dictionary length MAX_NB_WORDS
  • Training will take around an hour and trained model will be saved in Saved_model folder .
  • Accuracy and Loss plot is generated at last.

• Prediction

  • Import trained model from Saved_model folder
  • Run predict.py file or predict.ipynb Notebook.
  • New file solution.csv will be created in root folder.

Datasets is available at https://drive.google.com/drive/folders/1hOpQ2LpKixp3sj3CGv9700ETIp2tUkFR

Data for competition should be in the form of csv file at 'data' folder in root Directory

The solution file should be named solution.csv and the format is the same as sample.csv