meta silvae

meta silvae (latin for "beyon forest" and acronym of Magister Efficiens Temperat Arbore silvae, which translates to "The efficient master mixes trees of the forest", but also "The efficient master mixes trees through silva) is an abstract interpretation based tool for training decision tree classifiers, in particural we aim at enhancing stability-related properties of classifiers.

meta silvae deploies a genetic algorithm to evolve a poluation of decision trees, and internally uses silva in order to assert fitness of individual trees in terms of accuracy and stability.

Requirements

• Any C99-compatible C compiler

Installation

To install meta silvae you need to clone or download the source code files from this repository and compile them. There are no additional requirements nor dependencies:

git clone https://github.com/abstract-machine-learning/meta-silvae

or:

wget https://github.com/abstract-machine-learning/meta-silvae/archive/master.zip
unzip master.zip

then: cd meta-silvae/src make make install The executable file will be available under silva/bin/meta-silvae.

Every piece of code is documented using Doxygen. If you have Doxygen installed and wish to generate the documentation pages (HTML), run:

cd meta-silvae/src
make doc

Documentation will be available under meta-silvae/doc/html/index.html.

Usage

Run meta-silvae without arguments for a quick online help message. Full syntax is

bin/meta-silvae <training set> <output> [options]

Mandatory arguments:

• training set Path to training set file
• output Path to output file, silva format

Optional arguments:

• --fitness Fitness function used to estimate performance of a tree
  • linear Linear combination of standard performance indicators
• --select Criterion used to choose parents for crossover
  • uniform Choses a parent with a uniformally distributed probability
  • roulette-wheel Uses standard Roulette Wheel algorithm
• --crossover Crossover function
  • one-point Substitutes a subtree from first parent with a subtree from second parent
• --mutation Mutation function
  • none No mutation
  • grow Allows trees to grow
  • Z Allows both growth and pruning
• --mutation-base-probability p Base mutation probability in [0; 1], passed to mutation probability function
• --mutation-probability Mutation probability function
  • constant Always returns mutation base probability
  • encourage-variance Returns higher probabilities when variance is low
• --population-initial-size n Number of trees in initial population
• --population-max-size n Maximum number of trees in the population
• --population-next-size Function which computes size of next population
  • constant Always returns same size
• --population-generator Generator of intial population
  • blank Generates single-node trees (every sample in root)
  • from-forest Reads trees from a forest file
• --max-iteration n Maximum number of iterations for genetic algorithm
• --elitism n Number of best individual to copy to next iteration
• --split-search-aggressiveness p Fraction of splits to consider during split generation, in [0; 1]
• --allowed-features Features allowed for splits
  • all Uses every feature
  • uniform n Randomly choses n features from a uniform distribution
• --seed n Seed for random number generation

Data set format

See dedicated section on our data-collection repository, from which you can also download some ready-to-use datasets and models.