Rapid tutorial for a few machine learning techniques

These are a few jupyter notebooks introducing basics of supervised learning, generative modelling, and reinforcement learning. We discuss basic examples implemented in

• Python with JAX and Flax
• Julia using the Flux package.

Tutorials

• Supervised learning to classify the MNIST hand-written digits data set.
• Unsupervised learning to learn writing MNIST-like digits.

Physics examples

• Supervised learning to identify the phase transition of the classical Ising model

Set up Python environment

If you have anaconda installed, you can set up an environment with all required packages as follows:

1. Install Jupyter notebook and the nb_conda_kernels package in the base environment

    conda install -c conda-forge notebook
    conda install -c conda-forge nb_conda_kernels

2. Create a new environment

    conda create --name <insert_name_here> python=3.7 pip ipykernel ipywidgets tensorflow tensorflow-datasets matplotlib

3. Activate the new environment

    conda activate <insert_name_here>

4. Install JAX and Flax

    pip install jax jaxlib flax

5. Leave the new environment

    conda deactivate

Now you can open Jupyter notebook from the command line (with your base environment active) by entering

    jupyter notebook

In the notebook you should be able to choose a kernel named Python [conda env:<insert_name_here>]. When running this kernel, the libraries you just installed should be known.

Source: How to set up jupyter notebook to work with anaconda environments: https://towardsdatascience.com/how-to-set-up-anaconda-and-jupyter-notebook-the-right-way-de3b7623ea4a