ADL HW1 README

1. training

   • mkdir data/ and put all json in data

   • generate data

     • getSeq2Tag.py [training data jsonl path]
     • getSeq2Tag.py [validating data jsonl path]
     • getSeq2Tag.py [testing data jsonl path]
     • getSeq2Seq.py [training data jsonl path]
     • getSeq2Seq.py [validating data jsonl path]
     • getSeq2Seq.py [testing data jsonl path]

   • make folder to save checkpoint and chart

     • mkdir figplot checkpoint
     • mkdir figplot_seq2seq checkpoint_seq2seq
     • mkdir figplot_attentionBi checkpoint_attentionBi

   • seq2tag

     • python3.6 trainSeq2Tag.py [training data path] [validating data path] [embedding path] [ground truth jsonl path]

   • seq2seq

     • python3.6 trainSeq2Seq.py [training data path] [validating data path] [embedding path] [load model ('none')]

   • attention

     • python3.6 trainAttention.py [training data path] [validating data path] [embedding path] [load model ('none')]

2. predicting

   • seq2tag
     • ./extractive.sh test.jsonl predict.jsonl
   • seq2seq
     • ./seq2seq.sh test.jsonl predict.jsonl
   • attention
     • ./attention.sh test.jsonl predict.jsonl

3. ploting

   • visualize attention

     • python3.6 plotingAttention.py [encoder path] [decoder path] [embedding path] [testdata path]

   • plot relative position of summary

     • python3.6 plotingRelative.py [seq2tag model path] [embedding path] [testdata path]