ECF - Current features


General:

- minimal config: only individual structure (genotypes used), all other components may be used by default
- XML configuration
- parametrized framework components (no recompilation needed)
- optional use of smart pointers (boost::shared_ptr)
- evolution state checkpointing and restoring (milestone files)
- batch (repeated) running with summary statistics
- adding user defined components: genotypes, algorithms, operators
- multilevel logging


Genotypes (individual structure):

- individuals may contain _any combination_ of genotypes (including user defined)
- tree (genetic programming, more details below)
- binary encoded real values (bounds, precision, dimension; 7 crx and 2 mutation operators)
- floating point (bounds, dimension; 14 crx and 1 mutation operators)
- bit string (2 crx and 2 mutation operators)
- permutation vector (13 crx and 3 mutation operators)
- planned: cartesian GP, GEP


Algorithms:

- most algorithms are _genotype independent_ - can be used with any combination of genotypes
- GA/GP flavor: steady-state tournament, roulette wheel, elimination (generation gap)
- clonal selection, immune optimization
- evolutionary strategy (ES)
- artificial bee colony (ABC)
- differential evolution (DE)
- particle swarm optimization (PSO)
- genetic annealing (simulated annealing in parallel)
- random search


Evolutionary environment:

- examples covering each genotype
- many mutation and crossover operators for each genotype
- individuals may contain multiple genotypes of any kind (+ used defined)
- panmictic (global) or distributed population
- 5 selection operators (fitness proportional, inverse fit. prop., best, worst, random)
- 5 termination operators (best solution stagnation, fitness value, max. generations, max. time, max. evaluations)
- parallel algorithms (GA/GP variants)
- algorithm independent distributed evolution (e.g. distributed PSO, DE)
- high-level algorithm implementation, genotype independent


Genetic programming:

- 5 crossover and 7 mutation operators
- predefined functions (math, boolean)
- no recompilation needed for function and terminal set redefinition (stated in config file)
- constants, ERC terminals of different types
- user defined functions 
- user defined terminal types (STGP planned)
- multiple tree individuals (ADFs manual)


Parallel environment:

- parallel execution with MPI (multiprocessor node, network environment, cluster/grid)
- many parallelization options (global/distributed EC, sync/async, migration) regardless of the algorithm
- implicit parallelization concept