Precipitation-Nowcasting

This is an easy-to-understand implementation of ConvLSTM model(fisrt proposed by [Xinjian Shi et al.])(https://arxiv.org/abs/1506.04214https://arxiv.org/abs/1506.04214) in a real-world precipitation nowcasting problem with Pytorch. Here presents the guidance on how to run this project by yourself. Have fun!

DATA

Two open-sourced datasets are available for training and testing in this project.

1. A pre-masked radar datasets.(Included in the package)
2. Tianchi CNKI 2017 dataset（Provided by Shenzhen Meteorological Bureau）.This dataset is not included yet. However, You can download the datasets here

Getting Started

Prerequisites

Environment:

• Ubuntu 16.04+
• Anaconda 3-5.1
• Python 3.6
• CUDA 8

Python 3.6 Packages needed:
  Arrow   Pytorch 0.3.1   PIL

Installing

1. Install CUDA8

  \# sudo dpkg -i cuda-repo-ubuntu1604-8-0-local-ga2_8.0.61-1_amd64.deb
  \# sudo apt-get update
  \# sudo apt-get install cuda

2. Download and install Anaconda environment

  \# wget https://repo.continuum.io/archive/Anaconda3-5.1.0-Linux-x86_64.sh
  \# bash Anaconda3-5.1.0-Linux-x86_64.sh

    More on how to install Anaconda, see this passage

3. Install an environment(optional but suggested)

  \# conda create -n pytorch python=3.6 

4. Activate your new-built environemt and install Pytorch and torchvision

  \# source activate pytorch 
  \# pip3 install torch torchvision

Train the model

1. Download the all package and unpack it with the command:

  \# tar zxvf conlstm.tar.gz YOUR WORK PATH

Note: you also need to unpack the files in the original data directory before training

2. Train the model

  Python3 training.py

Running the test

Run the test.py with the command.

  python3 test.py  

Evaluate your model's performance by running

  python3 evaluate.py

Authors

 cxxixi
 pqx

Notes

1. Notes on ConvLSTM