For Kaggle competition: https://www.kaggle.com/c/recursion-cellular-image-classification Background information: https://www.rxrx.ai/#the-data
The full Dataset for this project is available here: https://www.kaggle.com/c/recursion-cellular-image-classification/data Download and extract the full dataset to this directory - you should have a test and train folder with subfolders and images.
Keras sklearn numpy pandas Tensorflow-gpu cv2 imgaug wandb (for logging and analysis)
Make sure the above libraries are available. Using wandb is highly encouraged since training variables can
easily be loaded from the config-defaults.yml
file. However, it can be run by configuring the TrainingRunner
in train-wb.py with the desired options.
To run the project using wandb, run this command from the command line after logging in with your wandb account (its free - sign up at www.wandb.com):
wandb run python train-wb.py
The CSV files have been preprocessed and re-saved for ease of training and testing. When using the multi-generator, use:
train_root.csv
for training full data
test_root.csv
for testing
train_controls_root.csv
for using training control data
all_controls_root.csv
for using all control data