Brain-Like-Computation-Project

Group project for NX-414 course at EPFL

Authors

• Ernesto Bocini: @ernestoBocini
• Florence Crozat @florence11

File structure

Report

• report.pdf: one page report of the project including explanations of methods and our results.

PY files:

• 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.

Notebooks:

• 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

Data folder:

• activationsResNet50: activations for week5
• cornet-useful: cornet weights and images. See also cornet models
• resnet50_improved: result submission

How to reproduce our results

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.

Create the environment

Make sure your environment satisfies the following fundamental requirements:

• Python 3.7+
• NumPy module
• PyTorch 1.13 module
• matplotlib module

Required packages

• 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

Run the code

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 .