DGMR-pytorch

An implementation of Deep Generative Model of Radars from DeepMind in PyTorch

Dependencies

• dask==2022.9.1
• matplotlib==3.5.1
• numba==0.55.1
• numpy==1.21.0
• pandas==1.4.1
• properscoring==0.1
• pytorch-lightning==1.5.10
• torchvision==0.11.3
• pytorch

Execute pip install -r requirements.txt to install required packages (except for torch).

For installing torch, please check out https://pytorch.org/get-started/locally/ to figure out the version that works on your device.

Workflow

Data preparation

• Step 1: Read your own data.
• Step 2: Store your data into numpy.ndarray with proper data type (e.g. int16, float64).
• Step 3: Make sure your data are sorted by time.
• Step 4: Note the array's storing data type and data shape, these information will be needed in config file.
• Step 5: Save the array with the format .dat or .npy. (No headers please!)

For more information, please check out numpy's documentation.

• Data types https://numpy.org/doc/stable/user/basics.types.html
• How to save array? https://numpy.org/doc/stable/reference/generated/numpy.save.html
• Acceptable data shape: (N, H, W)
  • N: number of rainfall images
  • H: height of rainfall images
  • W: width of rainfall images

How to run DGMR?

• Step 1: Prepare data (see Data Preparation).
• Step 2: Prepare rain records .csv file. (optional) If rain records are not prepared, it will be calculated automatically in our program. Example of csv:

index	nonzeros
0	number of nonzeros of image 0
1	number of nonzeros of image 1
...	number of nonzeros of images
N-1	number of nonzeros of image N-1

• Step 3: Prepare config file. Please see configs/README.md

• Step 4: Execute the code

  • train: python3 main.py -c /path/to/your/config -m train
  • validate: python3 main.py -c /path/to/your/config -m val
  • test: python3 main.py -c /path/to/your/config -m test

Tensorboard

Execute this command to see training results. tensorboard --logdir /path/where/records/are/stored