I am supervising the Leren & Beslsissen course of 2020. This repository functions as a provided-code hub for the students. It contains example models, dataloaders and training scripts in PyTorch, as well as, meeting notes and assignment instructions.

Currently there are two convnets for image-data (lenet5 and resnet), two recurrent-nets for text-data (rnn and lstm) and one fully connected (fc) for any data-type.

Currently contains MNIST, stanford-sentiment and an example toy dataset to show how you can make your own.

Is a class with a pre-made training script as an example for classification. If regression or something else is endeavoured, then it will need to be altered by student, it functions more as an example. Can be used like the shallow classes in sklearn where you just call model.fit(x, y). However, here it takes in as arguments at initialisation:

• the train-dataloader
• the test-dataloader
• optimizer
• model
• number of epochs
• loss-function
• device.

Thereafter you call .train() instead.

• Ensure you have pip3 installed

sudo apt install python3-pip

• Install virtualenv

sudo pip3 install virtualenv

• Create env

virtualenv -p python3 ~/virtualenvs/NAME --system-site-packages

• Activate

source ~/virtualenvs/NAME/bin/activate

• Install requirements

pip3 install -r requirements.txt

• If you use an IDE, in the settings set project-interpreter to:

~/virtualenvs/NAME/bin/python3.7

• Run

python3 example.py

• Example on colab