Description:

The Maja Machine Learning Framework (MMLF) is a general framework for problems in the domain of Reinforcement Learning (RL) written in python. It provides a set of RL related algorithms and a set of benchmark domains. Furthermore it is easily extensible and allows to automate benchmarking of different agents. Among the RL algorithms are TD(lambda), CMA-ES, EANT, Fitted R-Max, and Monte-Carlo learning. MMLF contains different variants of the maze-world and pole-balancing problem class as well as the mountain-car testbed.

Changes to previous version:

• Experiments can now be invoked from the command line
• Experiments can now be "scripted"
• MMLF Experimenter contains now basic module for statistical hypothesis testing
• MMLF Explorer can now visualize the model that has been learned by an agent

BibTeX Entry: Download

Supported Operating Systems: Agnostic

Data Formats: None

Tags: Reinforcement Learning, Optimization, Evolution, Toolbox, Neuroevolution

Archive: download here

Other available revisons

Version	Changelog	Date
1.0	Experiments can now be invoked from the command line Experiments can now be "scripted" MMLF Experimenter contains now basic module for statistical hypothesis testing MMLF Explorer can now visualize the model that has been learned by an agent	September 13, 2011, 15:13:56
0.9.10	Monitor can now also be configured in MMLF Explorer MMLF Experimenter can now also load results of a prior experiment MMLF Experimenter allows to specify whether world runs should be conducted sequentially or concurrently Added simple LinearMarkovChain environment Added "Writing an experiment" tutorial Removed EANT and CGE from MMLF	April 8, 2011, 15:19:56
0.9.9	Configuration files for worlds (i.e. for agents and environments) are now in the yaml syntax (and no longer in xml) MMLF GUI can now load and store world configuration files Experiments in the MMLF Experimenter can now consist of arbitrary many worlds (and not just two agents in the same world) MMLF Experimenter supports now parallel execution of experiments (each world in a separate process). This is based on the multiprocessing framework. Logging to files and providing information for the GUI is now handled both using Observables. This reduces code duplication. Added Actor-Critic agent Added Pinball Maze environment Added Discrete BRIO Labyrinth environment Added model for discrete environments which stores the transition probabilities for each (state, action, successor state) in a dict entry Added planner interface and refactored dynamic programming, prioritized sweeping etc. such that they implement this interface Added several viewers (both environment-specific as well as general purpose viewers for trajectories, value functions, policies etc.)	February 21, 2011, 14:05:02
0.9.8	Added "Experimenter" GUI to MMLF that allows to compare the performance of different agents Various small improvements of the GUI Added a simple two-player card game environment called "Seventeen and Four"	June 4, 2010, 11:39:55
0.9.7	Added model-based learning algorithms DYNA-TD and model-based direct policy search (MBDPS). Added an initial version of a graphical user interface	May 7, 2010, 21:24:11
0.9.6	Initial Announcement on mloss.org.	December 4, 2009, 08:14:56

Comments

No one has posted any comments yet. Perhaps you'd like to be the first?

Leave a comment

You must be logged in to post comments.