- Ernesto Bocini: @ernestoBocini
- Florence Crozat @florence11
- report.pdf: one page report of the project including explanations of methods and our results.
- test.py
- File that contains the best performing model. Use this file to reproduce our results or to test your dataset.
- utils.py
- File that contains various helper functions for the project.
- EDA.ipynb:
- file containing exploratory data analysis
- week5.ipynb:
- contains ridge regression, ridge regression with PCA and resnet50
- week6.ipynb:
- contains shallow CNN with optimization
- week7.ipynb
- contains best model and cornet models
- activationsResNet50: activations for week5
- cornet-useful: cornet weights and images. See also cornet models
- resnet50_improved: result submission
We assume that the repository is already downloaded and extracted, that the IT_data.h5 is downloaded and extracted in the data folder at the root of the program. We further assume that Anaconda is already installed.
Make sure your environment satisfies the following fundamental requirements:
- Python 3.7+
- NumPy module
- PyTorch 1.13 module
- matplotlib module
- Required packages for the best model:
- h5py
- os
- Image from PIL
- pickle
- resnet50, ResNet50_Weights from torchvision.models
- tqdm
- explained_variance_score from sklearn.metrics
- Required packages for running all the notebooks:
- all packages above
- optuna
From the root folder of the project
python test.py
Careful: training might be time consuming. The model has been trained and runned using the following machine:
- 16 vCPU, 104 GB di RAM, NVIDIA T4 x 1 .