Skip to content

Latest commit

 

History

History
88 lines (57 loc) · 1.68 KB

README.md

File metadata and controls

88 lines (57 loc) · 1.68 KB

This is an implementation of our paper in COLING2020:

One Comment from One Perspective: An Effective Strategy for Enhancing Automatic Music Comment Paper Path

This code is based on our baseline MMPMS: MMPMS

Requirements

  • Python == 3.7
  • PaddlePaddle == 1.8.1
  • NLTK == 3.4.5
  • numpy == 1.18.1
  • pandas == 1.0.3

Folder structure

/dataset                 : two datasets constructed by us
/data/ES_newdata/  	     : data path
/models  				 : models design
/output 				 : logs and results
run.py 					 : train or referece
eval.py 	   		     : eval the result of inference

Prepare data

The vocabulary and the preprocessed data will be saved in the folder:

data/ES_newdata/
├── music.train                need to prepare
├── music.test					 need to prepare
├── music.valid						 need to prepare
├── music.train.pkl				
├── music.valid.pkl
├── music.test.pkl
├── addcomments_small_embedding.txt        embedding file, need to prepare: https://ai.tencent.com/ailab/nlp/en/index.html
└── vocab.json

Preprocess the data by running preprocess.py :

python preprocess.py

Train

To train a model, run:

python run.py

Not use the distinction:

python run.py --fLoss_mode no

Use gpus:

python run.py --use_gpu True

Inference

python run.py --infer --model_dir MODEL_DIR --result_file RESULT_FILE

Evaluation

python eval.py RESULT_FILE