A library for constructing Neural Networks in Elixir or Erlang.
Neurlang is essentially a port of the DXNN2 core neural network elements.
However, unlike DXNN2, Neurlang:
- Is written in Elixir
- Is easily usable in non-TWEANN's (Topology WEight Adapting Neural Networks).
The goal is that any neural network should be able to be built using the components available in Neurlang. If you want to build a neural network, and Neurlang is missing something you need, please file an issue with a feature request.
Status: Neurlang is still in the process of being built, and is therefore is about as stable as the price of bitcoins
# Create nodes
neuron = Neuron.start_node(bias: 10, activation_function: function(identity/1))
sensor = Sensor.start_node(sync_function: fake_sensor_data([[1, 1, 1, 1, 1]]))
actuator = Actuator.start_node([])
# Wire up network
connect(from: sensor, to: neuron, weights: [20, 20, 20, 20, 20])
connect(from: neuron, to: actuator)
# tap into actuator for testing purposes
NodeProcess.add_outbound_connection(actuator, self())
# feed input into sensor
NodeProcess.sync(sensor)
# verify actuator output
assert actuator_next_output() == 110
Neurlang provides three types of neural elements, all of which are gen_server processes
which can link up with each other and exchange messages.
-
Sensor
-
Neuron
-
Actuator
DXNN2 - Pure Erlang TWEANN (Topology WEight Adapting Neural Network). Neurlang is essentially a port of the DXNN2 core neural network elements.
Handbook of Neuroevolution Through Erlang by Gene Sher.
One time:
dialyzer -DDIALYZER --output_plt .depsolver.plt --build_plt --apps erts kernel stdlib crypto public_key $ELIXIR/lib/elixir/ebin $ELIXIR/lib/ex_unit/ebin
where $ELIXIR points to your elixir install dir, eg, /usr/local/lib/elixir
Each time:
dialyzer --quiet --no_check_plt --plt .depsolver.plt -Wunmatched_returns -Werror_handling -Wrace_conditions ebin
or
dialyzer --quiet --no_check_plt --plt .depsolver.plt -Wunderspecs -Wunmatched_returns -Werror_handling -Wrace_conditions ebin