DeepMC

DeepMC architecture

• DeepMC(Kumar, Peeyush, et al. "Micro-climate Prediction-Multi Scale Encoder-decoder based Deep Learning Framework." Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining. 2021.) 논문의 모델을 구현하기 위해 만든 프로젝트

• WPD(Wavelet Packet Decomposition) output, 논문에서는 1 level이 long scale인데, 이 코드에서는 5 level이 long scale(아래 그림은 이 코드에서의 WPD 결과)

DeepMC 코드 구조

• Encoder
  • CNN-LSTM, CNN stacks
  • net/encoder.py
    • LSTMstack
    • CNNstack
• 2 level Attention layer
  • Position based content attention, Scaled guided attention
  • net/attention.py
    • Position_based_content_attention
    • Scaled_Guided_Attention
• Decoder
  • LSTM with FC-layer
  • net/decoder.py
    • Decoder
• All network
  • Encoder, 2 level attention, Decoder
  • net/deepmc.py
    • DeepMC

Library version

• python==3.8.5
• pytorch==1.8.1
• torchvision==0.9.1
• opencv-python==4.5.2.54
• pytorch-lightning==1.3.8

실행 방법

python trainer.py