Skip to content

Latest commit

 

History

History
60 lines (42 loc) · 1.53 KB

README.md

File metadata and controls

60 lines (42 loc) · 1.53 KB
Raw

3-Body with Neural Network

This is a proposed algorithm to solve 3-body problem using neural network used for the assignment in the course Parallel Computing. In this repo, we've already provided SIMPLE and EASY-to-handle command line tools, a pretrained model and preprocessed dataset.

Minimum Requirements

Disk Size: 35MB
RAM: 16GB
GPU Memory Capacity: 4 GB
CPU Memory Capacity: 16 GB
Operation System: Window 10 or Ubuntu 20.04 LTS

Note: If you use Linux, replace python as python3 when using the command line

Dependency

python==3.7.4
torch==1.7.1
torchvision==0.8.2
numpy==1.19.2
cupy-cuda101=8.1.0

How to use

Data Preparation (Optional)

  1. Generate the data:
    cd ANN
    python generate_data.py --data_size dataSize

Artificial Neural Network

Training (Optional)

  1. Once obtain the dataset, go to the ANN folder: cd ANN
  2. Train the model: python 3-body_ann.py --num_epochs num_epoch --saved_every_epoch period_to_save --batch_size batch_size

Inference

Test accuracy of the model

  1. Go to the ANN folder: cd ANN
  2. Use following command: python 3-body_ann_test_accuracy.py --model model_path

Test speed of the model

  1. Go to the ANN folder: cd ANN
  2. Use following command: python 3-body_ann_test_speed.py --model model_path --end_time endTime

Normal Parallel Implementation

Test speed of NumPy implementation

python 3-body_numpy.py --end_time endTime

Test speed of CuPy implementation

python 3-body_cupy.py --end_time endTime