Question Classifier

Requirements

• edict
• pytorch

References

• Text preprocessing file was used from one of my previous projects which was borrowed from another repository RonghangHu-CMN

Dataset

• CoQA a dialogues based dataset. Not questions - story sentences and questions - the one greater than 4 words.
• Number of training samples : 191367, Validation samples : 13638
• Basic preprocessing done in read_data.py

Implementation details

• Maximum words - 15, padded with zero
• GloVe 300d with 72704 words
• Finetuned embedding
• Batch size - 10

Model

• Vanilla LSTM model, No attention - 99% train and val accuracy.
• 1 layer - unidirectional
• A dropout rate of 0.4

Results

• The model gets trained pretty well in 1 epoch - upto 99% accuracy. I've set the max number of epochs as 2