Always sparse. Never dense. But never say never. A repository for the Adaptive Sparse Connectivity concept and its algorithmic instantiation, i.e. Sparse Evolutionary Training, to boost Deep Learning scalability on various aspects (e.g. memory and computational time efficiency, representation and generalization power).

Stars: ✭ 182 (+1200%)

Mutual labels: neuroevolution

View All Similar Projects ➔

apxr_run

A topology and parameter evolving universal learning network originally created by Gene Sher.

Requirements

More information on installing Erlang/OTP & Rebar3 can be found here.

Introduction

apxr_run is a distributed topology and weight evolving artificial neural network originally created by Gene Sher. Significant changes are a deeper integration of the OTP application structure, replacement of the single scape process with a quad tree, converting the sensor -> scape and actuator -> scape processes to be asynchronous, dropping Mnesia in favour of RocksDB and initial work to integrate with Elixir.

How to:

1. Build

$ rebar3 compile

2. Run Dialyzer (Erlang static analysis tool)

$ rebar3 dialyzer

3. Run Eunit (Unit tests)

$ rebar3 do eunit, cover --verbose

4. Run Xref (Cross reference analysis)

$ rebar3 xref

5. Run

$ rebar3 shell
1> experiment_mgr:run().

6. Generate release

$ rebar3 release -d -p priv/lib/elixir/ebin/

7. Run release

$ _build/default/rel/apxr_run/bin/apxr_run console

8. Observe

1> observer:start().

1> observer_cli:start().

9. Conduct an experiment

The experiment_mgr process composes experiments by performing N evolutionary runs, and then produces statistical data and chart ready files of the various evolutionary dynamics and averages.

All these files can be found under the /experiments directory.

Configuration

Every application of the system needs a corresponding configuration file. The /config directory contains a number of examples that can be used to get started.

Remember, we do not need to force the system to use any one particular approach. We can set it in the constraints to use all available functionality and the system will evolve it all.

Benchmarks

Three benchmarks are included:

Double pole balancing without damping
Double pole balancing with damping
Discrete T-maze

See the /doc/examples/benchmarks.md for more information.

Applications

Two applications are included:

Flatland
FX

See the /doc/examples/applications.md for more information.

Documentation

Developers

The most useful resource for developers working on this project will be the inline documentation.

General

For a more general introduction, see the /docs directory for a quick introduction to neuroevolution and related topics.

For a more thorough introduction, the book "Handbook of Neuroevolution Through Erlang" by Gene Sher is highly recommended. Indeed, much of the content in the Docs section is taken from this book as is the majority of the inline documentation found in the source code. So, credit for the good stuff goes to him.

Conventions & rules

Inaka's erlang_guidelines should be used as the basis for things that may be used as a reason to reject a PR.

Related publications

Handbook of Neuroevolution Through Erlang by Gene Sher.
Agent-Based Modeling Using Erlang by Gene Sher.

Note that the project description data, including the texts, logos, images, and/or trademarks, for each open source project belongs to its rightful owner. If you wish to add or remove any projects, please contact us at [email protected].

Cheap and reliable Node.js hosting starts at $3/month, and $1/month static HTML hosting

Rober-t / apxr_run

Programming Languages

Labels

Projects that are alternatives of or similar to apxr run