Statistics and Machine Learning in Python

• pdf
• www

Structure

Courses are available in three formats:

1. Jupyter notebooks.

2. Python files using sphinx-gallery.

3. ReStructuredText files.

All notebooks and python files are converted into rst format and then assembled together using sphinx.

Directories and main files:

introduction/
├── machine_learning.rst
└── python_ecosystem.rst

python_lang/                        # (Python language)
├── python_lang.py # (main file)
└── python_lang_solutions.py

scientific_python/
├── matplotlib.ipynb
├── scipy_numpy.py
├── scipy_numpy_solutions.py
├── scipy_pandas.py
└── scipy_pandas_solutions.py

statistics/                         # (Statistics)
├── stat_multiv.ipynb               # (multivariate statistics)
├── stat_univ.ipynb                 # (univariate statistics)
├── stat_univ_solutions.ipynb
├── stat_univ_lab01_brain-volume.py # (lab)
├── stat_univ_solutions.ipynb
└── time_series.ipynb

machine_learning/                   # (Machine learning)
├── clustering.ipynb
├── decomposition.ipynb
├── decomposition_solutions.ipynb
├── linear_classification.ipynb
├── linear_regression.ipynb
├── non_linear_prediction.ipynb
├── resampling.ipynb
├── resampling_solution.py
└── sklearn.ipynb

optimization/
├── optim_gradient_descent.ipynb
└── optim_gradient_descent_lab.ipynb

deep_learning/
├── dl_backprop_numpy-pytorch-sklearn.ipynb
├── dl_cnn_cifar10_pytorch.ipynb
├── dl_mlp_mnist_pytorch.ipynb
└── dl_transfer-learning_cifar10-ants-

Build

After pulling the repository execute Jupyter notebooks (outputs are expected to be removed before git submission).

make exe

Build the pdf file (requires LaTeX):

make pdf

Build the html files:

make html

Clean everything and strip output from Jupyter notebook (useless if you installed the nbstripout hook, ):

make clean

Dependencies

The easier is to install Anaconda at https://www.continuum.io with python >= 3. Anaconda provides

• python 3
• ipython
• Jupyter
• pandoc
• LaTeX to generate pdf

Then install:

1. sphinx-gallery

pip install sphinx-gallery

2. nbstripout

conda install -c conda-forge nbstripout

Configure your git repository with nbstripout pre-commit hook for users who don't want to track output in VCS.

cd pystatsml
nbstripout --install

3. Git LFS for datasets

a. Install Git LFS

git lfs install

b. select the file types you'd like Git LFS to manage

git lfs track "*.npz"
git lfs track "*.npy"
git lfs track "*.nii"
git lfs track "*.nii.gz"
git lfs track "*.csv"

b. Now make sure .gitattributes is tracked:

git add .gitattributes

4. LaTeX (optional for pdf)

For Linux debian like:

sudo apt-get install latexmk texlive-latex-extra

5. MS docx (optional)

docxbuilder

a. Install

pip install docxbuilder
pip install docxbuilder[math]

b. Build

make docx