Representation learning: propositionalization and embeddings

This repository is a growing collection of supplementary material for our book Representation learning: propositionalization and embeddings. Currently, a number of Jupyter notebooks illustrating the selected parts of the book is available.

Table of contents

1. Introduction to Representation Learning
2. Machine Learning Background
3. Text Embeddings
   1. LSA and LDA
   2. word2vec
   3. BERT
4. Propositionalization of Multi-Relational Data
   1. Wordification
   2. python-rdm
5. Graph and Heterogeneous Network Transformations
   1. node2vec
   2. metapath2vec
   3. hinmine
6. Unifying approaches
   1. StarSpace
   2. propDRM

How to use

Docker

The notebooks work best in a local installation containing Jupyter lab and other required packages. If you prefer docker or if you experience difficulties running the notebooks on your host operating system, you can try using the provided docker-compose.yml file as follows:

git clone [email protected]:vpodpecan/representation_learning.git
cd representation_learning
docker-compose up

When the container is up and running it will return a link to the Jupyter environment such as http://127.0.0.1:8888/?token=159090399d58b41041bfc812cf2bf5aa1779fb54a6170005. There you can open and run the provided notebooks.

Local installation

Requirements

• python 3.8+ (3.6 and 3.7 were also tested and should work as well)
• jupyterlab

In addition, each notebook has its own requirements which are installed when the notebook is executed for the first time.

Preparing the environment

1. Create and activate a virtual environment.

   • Linux

     python3 -m venv myEnv
     source myEnv/bin/activate

   • Windows

     python3 -m venv myEnv
     myEnv\Scripts\activate

2. Clone the repository

   git clone https://github.com/vpodpecan/representation_learning.git

3. Install and run jupyterlab. The following commands install jupyterlab and run it.

   pip install jupyterlab
   cd representation_learning
   jupyter lab

4. Open the link in a web browser and select a notebook.

How to contribute

Contributions are welcome! You are welcome to contribute corrections, new notebooks, examples, figures or any other material related to the contents of the book.

License

The code and materials in this repository are licensed under the MIT license except where stated otherwise.