Two hands-on workshops are planned on Wednesday and Thursday morning. Both of them will be based on Jupyter notebooks and Python packages.

If you do not have Jupyter notebook, you can install it using different options:

If you use conda:

conda install -c conda-forge notebook

or if you use pip:

pip install notebook

Another option will be to follow the tutorial on Google colab. This alternative can be a good solution if you want to attend the workshop, but you do not have a lot of experience with notebooks, python or you do not want to install the necessary dependencies on you machine. In this case, a link will be provided and all the codes and interactive demonstrations will run remotely without installing anything on your machine. Please note that you need a Google account to access Google colab.

Hands-on Deep Learning workshop (Nicolas Audebert)

This hands-on tutorial demonstrates the use of some machine learning toolboxes, more specifically scikit-learn and PyTorch. scikit-learn is a machine learning library for Python built on NumPy/SciPy that is both simple to use and fast.
PyTorch is a both a scientific computing library with native GPU acceleration and a set of tools for implementing and optimizing deep neural networks.

The following packages are needed and it is preferred that they are installed beforehand:

- numpy
- matplotlib
- seaborn
- pandas
- torch
- torchvision
- scikit-learn
- scikit-image
- skorch
- umap-learn
- hdbscan
- h5py

Depending on your local setup, these packages can be installed through conda or pip.

Some level of familiarity with Python and its scientific computing ecosystem (e.g. NumPy) is expected.

Hands-on Multi-scale Analysis workshop (J.-F. Robitaille)

This hands-on workshop will present a non-exhaustive demonstration of some python tools available to the astrophysical community to perform multi-scale and related analysis.

The demonstrations are based on mainly two packages:

• TurbuStat - https://github.com/Astroua/TurbuStat
• Pywavan - https://github.com/jfrob27/pywavan

TurbuStat aimes at facilitating comparisons between spectral line data cubes. Included in this package are several techniques described in the literature which aim to describe some property of a data cube. It also includes an impressive collection of statistical analysis techniques, such as the PDF, the power spectrum, the ∆-variance, the dendrogram, etc.

Pywavan has been developed by J.-F. Robitaille and it revolves mainly around one function which is dedicated to the wavelet power spectrum analysis of a map and its Multi-scale non-Gaussian Segmentation (MnGSeg). It also contains functions to perform the classical Fourier power spectrum analysis, generate fractal simulations and do basic data manipulation, such as cutting fits maps, beam convolution, etc.

Dependencies

Demonstrations will be applied on fractal simulations of the ISM and on real data.

Here are the necessary packages that you will need to run the demonstrations:

Plots and matrices

pip install matplotlib numpy

Dowload, manage and display fits data

pip install aplpy astropy

Turbustat package and its dependencies

pip install turbustat astrodendro sklearn statsmodels radio_beam

Pywavan package

pip install git+https://github.com/jfrob27/pywavan.git