Kaggle Invasive Species Monitoring
This repository is my workflow for Kaggle's Invasive Species Monitoring competition.
In this competition, we are given images with and without Hydrengea, a species which is invasive in North America.
Running the code
Setting up the environment
First, clone the repository
git clone https://github.com/tetelestia/kaggle-invasive-species
Move into the new directory
cd kaggle-invasive-species
Create a virtual environment with Python 3 (3.5 or greater)
virtualenv -p python3 invasive_env
Then, activate the virtualenv and install all required packages
source invasive_env/bin/activate
pip install -r requirements.txt
Finally, PyTorch must be installed from the website. Go to pytorch.org and follow the installation instructions on their website.
If you are running on an Nvidia GPU (highly suggested), CUDA must be installed in order to use the GPU, and while you're at it, installing cuDNN is a good idea, as it decreases computation time by about a factor of two.
Directory structure
The data is to be set in folders as such:
- data
- train/*
- test/*
- sample_submission.csv
- train_labels.csv
Training
The training process begins by running train.py in the network directory. Currently, all that it does is trains a Resnet model of your choice, with no validation. More to come...