Types of Modules¶
Organisms¶
For more detailed information on organisms, see the Organisms page.
An organism is an individual agent and the target of evolution in MABE2. The genetic material of an organisms is stored in the genome. Organisms use brains to process input and determine outputs. Genomes and brains can be uniquely combined at runtime to easily create custom organisms, instead of creating them from scratch.
Genomes¶
For detailed information on genomes, see genome page link.
A genome is a source of heritable and mutable data. Genomes usually provide data that is used to construct other elements of a module. For example, a genome may be used by a brain to determine that brain’s structure or by a world to determine the properties of an agent’s body. The genomes interface requires access functions to write values into the genome (including randomizing the genome) and read values from the genome. Each type of genome defines its own internal data structures, access function behaviors, and mutation operators.
Brains¶
For detailed information on brains, see brain page link.
Brains are data processors that receive input and deliver output. Brains are the most common method by which agents communicate with worlds (although worlds interfacing directly with genomes is an option). Under this abstraction, genetic programs, artificial neural networks, Markov Brains, ~et cetera~ are classified as brains. Depending on the user’s context, they may find it easier to think of brains as controllers, solvers, solutions, I?O machines, or even as chemical processes. The brian interface specifies how brains must define its internal workings, including update function, internal states, and internal data structures.
Evaluators¶
For detailed information on genomes, see the evaluator page .
Evaluators are the functions on which organisms get tested. Evaluation modules are categorized by the types of IO they work with.
Static Evaluators¶
Static evaluators do not take any input. They produce output as the solution to a single problem that does not change
(typical of most evolutionary computation systems). The two types of static output are BitVector (more information
here ) and emp::vector<double>.
Value I/O¶
These evaluation modules provide a set of doubles (emp::vector<double> or
std::unordered_set<size_t,double>) as input, and expect a new vector of
doubles back as an output.
In some cases, these are used for choosing a move in a game, where the output vector lists all possible moves and the highest value of a legal move is the one that is automatically chosen. Such a system allows the organism to simply weight each possible move without needing to actually select one.
At the current moment there are no static evaluators.
Selectors¶
For detailed information on selectors, see selectors page link.
Selectors manage populations, select parents, oversee re- production, and terminate agents who get to be too old or fail to meet certain criteria. Roulette selection, tournament selection, lambda+n, lexicase selection, and MAP-Elites are all examples of the types of algorithms optimizers typically implement.
Placement¶
For detailed information on placement, see placement page link.
Placement modules dictate what happens when a new organism is placed into a population. This applies to both injected organisms, as happens at the start of the run, and offspring organisms.
Schema¶
For detailed information on schema, see schema page link.
Schema modules are those which don’t fit elsewhere. This includes modules such as MovePopulation, which dictates how organisms
are shifter from one population to another, and Mutate, which allows for extra mutations in a population.
Analyze¶
For detailed information on analysis modules, see analyze page link.
Analyze modules collect, process, and store population data.
Interface¶
For detailed information on interface, see interface page link.
- Interface modules allow the user to interact with and understand MABE objects and experiments. These modules handle
the I/O stream and generate output files.