Jupyter notebooks for the 'deep learning for practitioners' course

Dependencies

• Python (>= 3.11)
• notebook (>=7.3.2)
• numpy (>= 2.1.2)
• matplotlib (>=3.10.1)
• pandas (>=2.2.3)
• scikit-learn (>=1.6.1)
• torch (>= 2.7.0)
• tqdm (>=4.67.19)
• ipywidgets (>= 8.1.5)
• seaborn (>=0.12.2)

In the repo you find also a *yml file that can be used to initialize a conda-environment.

Install required packages

The best way to install the required packages is to use miniconda. You can get the actual and also older version directly from the anaconda website.

After the installation of miniconda, create a new environment, for example with:

conda create -n microcredential

Here, microcridential is the name of the example environment

Now activate the environment with:

conda activate microcredential

Please make sure that the environment is active throughout the installation process of the packages and while working with the notebooks.

First, install Python:

conda install python

If you want, you can specify the Python version you want to use. For example:

conda install python=3.11

installs Python version 3.11.

Together with Python, conda also installs pip in the environment, what we use to install the other necessary packages.

As PyTorch has to be installed either in a CPU only version, on in a GPU version (which also depends on your CUDA driver) we will first install all other packages mentioned above. This can be done with one line:

pip install notebook numpy seaborn pandas scikit-learn tqdm ipywidgets

Pip will automatically install all other necessary Python packages.

If all packages have been successfully installed, proceed to install PyTorch.

Install PyTorch

Check your operating system, decide if you want to run it on CPU only or with GPU support, and for the latter also check your CUDA version. Then go to the PyTorch website and select under START LOCALLY the conditions you want to use to install PyTorch. It will then generate the pip command to install PyTorch and all necessary packages.
NOTE Did not use the pip3 command, use pip instead.

Here is an example pip line to install it with CUDA 12.8:

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128

If you want to use pretrained models from timm, you can install it with:

pip install timm

Pre-defined environments

We added two pre-defined environments: one for linux, one for windows (without PyTorch).
You can create your conda environment also based on this:

conda env create -f environment_*.yml

Notebooks

1. Qickstart to PyTorch

2. Deep neural networks

   2.1. Introduction to multilayer perceptron (MLP)

   2.2. Introduction to convolutional neural networks (CNN)

   2.3. CNN with Imagenette

   2.4. ResNet

   2.5. YOLO

3. Evaluation and Datasets

   2.1. Evaluating Neural Network Performance: Metrics and Visualizations

   2.2. Custom Datasets

4. Vision Transformer

   4.1. Vision Transformer (ViT)

   4.2. Compact Convolution Transformer (CCT)

5. Self-supervised learning (SSL)

   6.1. Simple framework for Contrastive Learning of visual Representations (SimCLR)

   6.2. Bootstrap Your Own Latent (BYOL)

   6.3. Self-distillation with no labels (DINO)