CudaStonks

This project was developed for the Spring 2024 class of CS147 (GPU programming) at UCR.

It contains an hand written back propagating neural network for predicting stocks data, both a CPU-based implementation using plain python and numpy, and a GPU-based implementation using Numba.

There is also a PyTorch version for comparison.

Performance of the model on the S&P500

Setup

First of all install the required python packages:

# You can setup a virtual environment if you want
python3 -m pip install venv
python3 -m venv .venv
source .venv/bin/activate

# Install packages
python3 -m pip install -r requirements.txt

Run the code

Change directory to src and from there you can run the targets of the makefile.

• Run make cpu to run the cpu version of the neural network
• Run make gpu to run the gpu version of the neural network
• Run make pytorch to run the model on pytorch

Run on gpu

The code can be run on GPU provided that:

• You have a NVIDIA GPU
• You have cuda installed in your system

Links

• Presentation: https://docs.google.com/presentation/d/1pMCIn6s4FMmayNxdV6SKRwrzRtKbEc6xfHKGRCjdnIM/edit?usp=sharing
• Report: https://docs.google.com/document/d/1CZPaI57Etz3DFJPeZs57x3tjNPrIG3cxlGybwlbjN1g/edit?usp=sharing
• Colab: https://colab.research.google.com/drive/1C607n23h6gnq2w5iezJEZ5IX1nj7DIPI?usp=sharing