EOgmaNeo

Introduction

Welcome to the EOgmaNeo library!

EOgmaNeo is Ogma Corp's embedded and event based version of OgmaNeo

It is our primary and preferred implementation of Sparse Predictive Hierarchies, a fully online sequence predictor. This is no longer the case, please use OgmaNeo2.

EOgmaNeo currently runs exclusive on the CPU, unlike OgmaNeo. However, for most tasks, it is much faster. It also performs better in terms of end-result.

EOgmaNeo performs some optimizations not yet present in OgmaNeo, resulting in a massive speed boost. For example, on weaker hardware such as the Raspberry Pi it will run happily at 60FPS with ~10,000,000 synapses.

We used this software to build a small self-contained online-learning self driving model car: Blog post

The advantage of our software over standard Deep Learning techniques is primarily speed. A single Raspberry Pi 3 is enough to run simulations of networks with tens of millions of synapses at high framerates, while Deep Learning is often restricted to offline training on very large and expensive GPUs.

Bindings to Python are also included. The binding API approximately mimics the C++ API. Refer to README.md files in each subdirectory to discover more about each binding, and how to build and use them.

Overview

For a more detailed introduction, see OVERVIEW.md

EOgmaNeo is a fully online learning algorithm, so data must be passed in an appropriately streamed fashion.

The simplest usage of the predictive hierarchy involves calling:

    // Compute system
    eogmaneo::ComputeSystem cs(4); // Number of threads to use, e.g. CPU Core count

    eogmaneo::Hierarchy h;

    // Layer descriptors
    std::vector<eogmaneo::LayerDesc> lds(6);

    // Layer size
    const int layerWidth = 4;
    const int layerHeight = 4;
    const int layerColumnSize = 32;

    for (int l = 0; l < lds.size(); l++) {
        lds[l]._width = layerWidth;
        lds[l]._height = layerHeight;
        lds[l]._columnSize = layerColumnSize;
        // ...
    }

    // Create the hierarchy
    h.create({ { 2, 2 } }, { 16 }, { true }, lds, 1234); // Input width x height, input column size, whether to predict, layer descriptors, and seed

You can then step the simulation with:

    h.step(cs, sdrs, true); // Input SDRs, learning is enabled

And retrieve predictions with:

    std::vector<int> predSDR = h.getPredictions(0); // Get SDR at first (0) visible layer index.

Important note: Inputs are presented in a columnar SDR format. This format consists of a list of active units, each corresponding to a column of input. This vector is in raveled form (1-D array of size width x height).

All data must be presented in this form. To help with conversion, we included a few "pre-encoders" - encoders that serve to transform various kinds of data into columnar SDR form.

Currently available pre-encoders:

• ImageEncoder
• KMeansEncoder
• GaborEncoder

You may need to develop your own pre-encoders depending on the task. Sometimes, data can be binned into columns without any real pre-encoding, such as bounded scalars.

Requirements

EOgmaNeo requires: a C++1x compiler, and CMake.

The library has been tested extensively on:

• Windows using Microsoft Visual Studio 2013 and 2015,
• Linux using GCC 4.8 and upwards,
• Mac OSX using Clang, and
• Raspberry Pi 3, using Raspbian Jessie with GCC 4.8

CMake

Version 3.1, and upwards, of CMake is the required version to use when building the library.

Building

The following commands can be used to build the EOgmaNeo library:

mkdir build; cd build
cmake -DBUILD_PREENCODERS=ON ..
make

The cmake command can be passed the following optional settings:

• CMAKE_INSTALL_PREFIX to determine where to install the library and header files. Default is a system-wide install location.
• BUILD_PREENCODERS to include the Random and Corner pre-encoders into the library.

make install can be run to install the library. make uninstall can be used to uninstall the library.

On Windows it is recommended to use cmake-gui to define which generator to use and specify optional build parameters.

Examples

C++ examples can be found in the source/examples directory. Python, Java, and C# examples can be found in their sub-directories.

Refer to README.md files found in each sub-directory for further information.

Contributions

Refer to the CONTRIBUTING.md file for information on making contributions to EOgmaNeo.

License and Copyright

The work in this repository is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. See the EOGMANEO_LICENSE.md and LICENSE.md file for further information.

Contact Ogma via [email protected] to discuss commercial use and licensing options.

EOgmaNeo Copyright (c) 2017-2018 Ogma Intelligent Systems Corp. All rights reserved.