This repository contains an implementation of techniques related to Empirical Model Learning, a technique to enable Combinatorial Optimization and decision making over complex real-world systems. The method is based on the (relatively simple) idea of:
- Using a Machine Learning (ML) model to approximate the input/output behavior of a system that is hard to model by conventional means.
- Embedding such Empirical Model into a Combinatorial Optimization model.
The emphasis of EML is mostly on the techniques to perform the embedding. These should be designed so that the optimization engine can exploit the structure of the empirical model to boost the search process.
This repository includes:
- Actual embedding techniques (encodings and global constraints)
- Pre- and post- processing methods
- I/O support (in particular readers for popular ML libraries)
The EML technique has been originally designed by the Artificial Intelligence group of the University of Bologna, but the repository implements also contributions from other groups.
The repository has been just launched and currently contains only a fraction of the existing techniques related to EML. In particular, we have published the bare minimum to support the EML tutorial at IJCAI 2018, and the current documentation is very scarce. More content (and improved documentation) will be released in the coming months.
The EMLlib is part of the EML resources (including the EML tutorial), all available at http://emlopt.github.io.
pip install emllib
In details, every object in the EMLlib belongs to one of three groups:
- Core EML objects are the classes and functions that implement the main EML functionalities, such as encodings, pre- and post-processing methods
- Backends provide the interface the core EML objects with a specific optimization solver
- Readers allow to convert Machine Learning models built via popular frameworks in the internal formats used by the EMLlib
The repository is structured as follows:
- The
backendmodule contains all the available backends- The
basesubmodule defines the interface that all backends should implement - The remaining submodules provide solver-specifc implementations of the base API
- The
- The
netandtreemodules contains classes and functions to work respectively with Machhine Learning models, and in particular Neural Network and Decision Trees- The
describesubmodule contains classes and functions that are useful to represent ML models - The
embedsubmodule contains classes and functions that are useful to embed ML models into a combinatorial model - The
processsubmodule contains classes and functions that are useful to pre- and post-process ML models for use optimization - The
readersubmodule contains specific reader implementations
- The
- The
utilmodule contains classes and functions used by multiple core EML objects
The core EML library is implemented in pure Python (3.X) and relies on very few additional modules. However, for every practical task you will need to use at least one backend, and probably at least one reader.
In order to have the core EML functionalities working you will need to:
- Obtain a Python 3.X distribution
- Install numpy, e.g. via pip with
pip3 install numpy - Add the main folder of your repository to your python path. This can be done by setting the global
PYTHONPATHvariables to include the path to this repository
We are planning to add a pip based installation in the coming months.
A reader for Neural Networks built with the keras sequential API is provided in net.reader.keras_reader.
The reader itself does not require a specific installation, but you will need to install keras to make it work.
A reader for Decision Trees built with the scikit-learn is provided in tree.reader.sklearn_reader.
The reader itself does not require a specific installation, but you will need to install scikit-learn to make it work.
A backend for the CPLEX Mixed Integer Linear Solver is provided in backend.cplex_backend.
The backend itself does not require a specific installation, but you will need to install docplex to make it work.