A Physicist's Introduction to Machine Learning

Machine Learning PhD Course held at IFT (Madrid)

You can find the schedule in: https://moseranette.wixsite.com/doctorado/advanced-1

Contact email: [email protected]

Theory sessions delivered by Bryan Zaldivar. Email: [email protected]

Outline of the hands-on sessions

• Setup and basics (Days 11/03/2019 and 13/03/2019). Here
  • Installation party
  • Python basics
  • Numpy basics
• Classification (Day 20/03/2019). Here
  • Decision tree from scratch
  • Classification algorithms with sklearn
• Neural Networks (Day 27/03/2019). Here
  • Feed forward NN from scratch
• Unsupervised methods (Day 08/04/2019). Here
  • Gaussian Mixture Models and EM algorithm
  • Principal Component Analysis

Setup

We will teach Python using the Jupyter notebook, a programming environment that runs in a web browser. However, you can use any Python interpreter compatible with Jupyter notebooks.

Anaconda (Local, Load predefined env)

1. Download miniconda Python 3.7 (light version of Anaconda) from https://docs.conda.io/en/latest/miniconda.html

2. Installation instructions in https://conda.io/projects/conda/en/latest/user-guide/install/index.html#regular-installation

3. Update all the packages in conda, and itself, with: conda update --all

4. Setup the conda env from the env.yml file with: conda env create --file env.yml

   Depending on your conda version the syntax can be instead: conda create --file env.yml

5. Activate the environment with:

   • Windows: activate ML_Course
   • Linux, macOS: conda activate ML_Course or source activate ML_Course

6. Open Jupyter with: jupyter notebook

Anaconda (Local, Install your own packages)

1. Download miniconda Python 3.7 (light version of Anaconda) from https://docs.conda.io/en/latest/miniconda.html

2. Installation instructions in https://conda.io/projects/conda/en/latest/user-guide/install/index.html#regular-installation

3. Update all the packages in conda, and itself, with: conda update --all

4. Create a new environment with conda create -n ML_Course python=3.6

   Depending on your conda version the syntax can be instead: conda env create -n ML_Course python=3.6

5. Activate the environment with:

   • Windows: activate ML_Course
   • Linux, macOS: conda activate ML_Course or source activate ML_Course

6. Install the following packages with conda: conda install pip graphviz

7. Install the required packages with pip: pip install -r requirements.txt

8. Open Jupyter with: jupyter notebook

Google Colab (Cloud)

Google has released its own flavour of Jupyter called Colab, which has free GPUs available.

Here's how you can use it:

1. Open https://colab.research.google.com, click Sign in in the upper right corner, use your Google credentials to sign in.
2. Click GITHUB tab, paste https://github.com/davidggphy/2019_03_Curso_IFT and press Enter
3. Choose the notebook you want to open *.ipynb
4. Click File -> Save a copy in Drive... to save your progress in Google Drive
5. Optional: Click Runtime -> Change runtime type and select GPU in Hardware accelerator box

Warm-up

You can find extra information about how to install Anaconda and a tutorial on python basics in https://adgdt.github.io/2018-11-28-cftmat/.

Contributing

Contributions and issues are welcome. Thanks!

Issues

Graphviz executables not found. Windows. (Thanks to Eduardo García-Valdecasas)

1. Download and install graphviz-2.38.msi from https://graphviz.gitlab.io/_pages/Download/Download_windows.html
2. Set the path variable
   • Control Panel > System and Security > System > Advanced System Settings > Environment Variables > Path > Edit
   • add C:\Program Files (x86)\Graphviz2.38\bin