Skip to main content

NEGotiations Managed by Agent Simulations

Project description

.. image:: https://img.shields.io/pypi/pyversions/negmas.svg :target: https://pypi.python.org/pypi/negmas :alt: Python

.. image:: https://img.shields.io/pypi/status/negmas.svg :target: https://pypi.python.org/pypi/negmas :alt: Pypi

.. image:: https://img.shields.io/pypi/l/negmas.svg :target: https://pypi.python.org/pypi/negmas :alt: License

.. image:: https://img.shields.io/pypi/dm/negmas.svg :target: https://pypi.python.org/pypi/negmas :alt: Downloads

.. image:: https://img.shields.io/codacy/coverage/1b204fe0a69e41a298a175ea225d7b81.svg :target: https://app.codacy.com/project/yasserfarouk/negmas/dashboard :alt: Coveage

.. image:: https://img.shields.io/codacy/grade/1b204fe0a69e41a298a175ea225d7b81.svg :target: https://app.codacy.com/project/yasserfarouk/negmas/dashboard :alt: Code Quality

.. image:: https://img.shields.io/pypi/v/negmas.svg :target: https://pypi.python.org/pypi/negmas :alt: Pypi

.. image:: https://img.shields.io/travis/yasserfarouk/negmas.svg :target: https://travis-ci.org/yasserfarouk/negmas :alt: Build Status

.. image:: https://readthedocs.org/projects/negmas/badge/?version=latest :target: https://negmas/readthedocs.io/en/latest/?badge=latest :alt: Documentation Status

.. image:: https://img.shields.io/badge/code%20style-black-000000.svg :target: https://github.com/ambv/black

NegMAS is a python library for developing autonomous negotiation agents embedded in simulation environments. The name negmas stands for either NEGotiation MultiAgent System or NEGotiations Managed by Agent Simulations (your pick). The main goald of NegMAS is to advance the state of the art in situated simultaneous negotiations. Nevertheless, it can; and was used; in modeling simpler bilateral and multi-lateral negotiations, preference elicitation , etc.

.. note:: A YouTube playlist to help you use NegMAS for ANAC2019_ SCM_ league can be found here_

.. _ANAC2019: http://web.tuat.ac.jp/~katfuji/ANAC2019
.. _SCM: http://web.tuat.ac.jp/~katfuji/ANAC2019/#scm
.. _here: https://www.youtube.com/playlist?list=PLqvs51K2Mb8LlUQk2DHLGnWdGqhXMNOM-

Introduction

This package was designed to help advance the state-of-art in negotiation research by providing an easy-to-use yet powerful platform for autonomous negotiation targeting situated simultaneous negotiations. It grew out of the NEC-AIST collaborative laboratory project.

By situated negotiations, we mean those for which utility functions are not pre-ordained by fiat but are a natural result of a simulated business-like process.

By simultaneous negotiations, we mean sessions of dependent negotiations for which the utility value of an agreement of one session is affected by what happens in other sessions.

The documentation is available at: documentation_

.. _documentation: https://negmas.readthedocs.io/

Main Features

This platform was designed with both flexibility and scalability in mind. The key features of the NegMAS package are:

#. The public API is decoupled from internal details allowing for scalable implementations of the same interaction protocols. #. Supports agents engaging in multiple concurrent negotiations. #. Provides support for inter-negotiation synchronization either through coupled utility functions or through central control agents. #. The package provides sample negotiators that can be used as templates for more complex negotiators. #. The package supports both mediated and unmediated negotiations. #. Supports both bilateral and multilateral negotiations. #. Novel negotiation protocols and simulated worlds can be added to the package as easily as adding novel negotiators. #. Allows for non-traditional negotiation scenarios including dynamic entry/exit from the negotiation. #. A large variety of built in utility functions. #. Utility functions can be active dynamic entities which allows the system to model a much wider range of dynamic ufuns compared with existing packages. #. A distributed system with the same interface and industrial-strength implementation is being created allowing agents developed for NegMAS to be seemingly employed in real-world business operations.

To use negmas in a project

.. code-block:: python

import negmas

The package was designed for many uses cases. On one extreme, it can be used by an end user who is interested in running one of the built-in negotiation protocols. On the other extreme, it can be used to develop novel kinds of negotiation agents, negotiation protocols, multi-agent simulations (usually involving situated negotiations), etc.

Running existing negotiators/negotiation protocols

Using the package for negotiation can be as simple as the following code snippet:

.. code-block:: python

from negmas import SAOMechanism, AspirationNegotiator, MappingUtilityFunction
session = SAOMechanism(outcomes=10, n_steps=100)
negotiators = [AspirationNegotiator(name=f'a{_}') for _ in range(5)]
for negotiator in negotiators:
    session.add(negotiator, ufun=MappingUtilityFunction(lambda x: random.random() * x[0]))

session.run()

In this snippet, we created a mechanism session with an outcome-space of 10 discrete outcomes that would run for 10 steps. Five agents with random utility functions are then created and added to the session. Finally the session is run to completion. The agreement (if any) can then be accessed through the state member of the session. The library provides several analytic and visualization tools to inspect negotiations. See the first tutorial on Running a Negotiation for more details.

Developing a negotiator

Developing a novel negotiator slightly more difficult by is still doable in few lines of code:

.. code-block:: python

from negmas.sao import SAONegotiator
from negmas import ResponseType
class MyAwsomeNegotiator(SAONegotiator):
    def __init__(self):
        # initialize the parents
        super().__init__(self)

    def respond(self, offer, state):
        # decide what to do when receiving an offer
        return ResponseType.ACCEPT_OFFER

    def propose(self, state):
        # proposed the required number of proposals (or less) 
        pass

By just implementing respond() and propose(). This negotiator is now capable of engaging in alternating offers negotiations. See the documentation of Negotiator for a full description of available functionality out of the box.

Developing a negotiation protocol

Developing a novel negotiation protocol is actually even simpler:

.. code-block:: python

from negmas.mechanisms import Mechanism

class MyNovelProtocol(Mechanism):
    def __init__(self):
        super().__init__()

    def round(self):
        # one step of the protocol
        pass

By implementing the single round() function, a new protocol is created. New negotiators can be added to the negotiation using add() and removed using remove(). See the documentation for a full description of Mechanism available functionality out of the box [Alternatively you can use Protocol instead of Mechanism].

Running a world simulation

The raison d'être for NegMAS is to allow you to develop negotiation agents capable of behaving in realistic business like simulated environments. These simulations are called worlds in NegMAS. Agents interact with each other within these simulated environments trying to maximize some intrinsic utility function of the agent through several possibly simultaneous negotiations.

The situated module provides all that you need to create such worlds. An example can be found in the scml package. This package implements a supply chain management system in which factory managers compete to maximize their profits in a market with only negotiations as the means of securing contracts.

Acknowledgement

.. _Genius: http://ii.tudelft.nl/genius

NegMAS tests use scenarios used in ANAC 2010 to ANAC 2018 competitions obtained from the Genius_ Platform. These domains can be found in the tests/data and notebooks/data folders.

History

Release 0.3.2

  • updating dependencies to latest versions

Release 0.3.1

  • [Situated] Correcting multistage tournament implementation.

Release 0.3.0

  • [Situated] adding StatsMonitor and WorldMonitor classes to situated
  • [Situated] adding a parameter to monitor stats of a world in real-time
  • [Situated] showing ttest/kstest results in evaluation (negmas tournament commands)
  • [SCML] adding total_balance to take hidden money into account for Factory objects and using it in negmas tournament and negmas scml
  • [SCML] enabling --cw for collusion
  • [SCML] adding hidden money to agent balance when evaluating it.
  • [SCML] adding more debugging information to log.txt
  • [Situated] adding multistage tournaments to tournament() function
  • [Situated] adding control of the number of competitor in each world to create_tournament() and to negmas tournament create command
  • [Core] avoid invalid or incomplete outcome proposals in SAOMechanism
  • [Situated] adding metric parameter to evaluate_tournaments and corrsponding tournament command to control which metric is used for calculating the winner. Default is mean.
  • [SCML] adding the ability to prevent CFP tampering and to ignore negotiated penalties to SCMLWorld
  • [SCML] adding the possibility of ignore negotiated penalty in world simulation
  • [SCML] saving bankruptcy events in stats (SCML)
  • [SCML] improving bankruptcy processing
  • [SCML] deep copying of parameters in collusion
  • [Situated] saving extra score stats in evaluate_tournament
  • [Core] avoiding a future warning in pandas
  • [Situated] more printing in winners and combine commands
  • [Situated] removing unnecessary balance/storage data from combine_tournament_stats
  • [Situated] adding aggregate states to evaluate_tournament and negmas tournament commands
  • [Situated] adding kstest
  • [Situated] adding and disabling dependent t-tests to evaluate_tournament
  • [Situated] adding negmas tournament combine to combine and evaluate multiple tournaments without a common root
  • [Situated] avoiding an exception if combine_tournament is called with no scores
  • [Situated] always save world stats in tournaments even in compact mode
  • [SCML] reversing sabotage score
  • [SCML] correcting factory number capping
  • [SCML] more robust consumer
  • [Core] avoid an exception if a ufun is not defined for a negotiator when logging
  • [SCML] controlling number of colluding agents using --agents option of negmas tournament create
  • [SCML] changing names of assigned worlds and multiple runs to have a unique log per world in tournament
  • [SCML] controlling warnings and exception printing
  • [SCML] increasing default world timeout by 50%
  • [SCML] removing penalty processing from greedy
  • [Core] avoid negotiation failure for negotiator exceptions
  • [SCML] correcting sabotage implementation
  • [CLI] adding winners subcommand to negmas tournament
  • [CLI] saving all details of contracts
  • [CLI] adding --steps-min and --steps-max to negmas tournament create to allow for tournaments with variable number of steps
  • [CLI] removing the need to add greedy to std competition in anac 2019
  • [CLI] saving log path in negmas tournament create
  • [CLI] removing errroneous logs
  • [CLI] enabling tournament resumption (bug fix)
  • [CLI] avoiding a problem when trying to create two tournaments on the same place
  • [CLI] fairer random assignment
  • [CLI] more printing in negmas tournament
  • [CLI] using median instead of mean for evaluating scores
  • [CLI] Allowing for passing --world-config to tournament create command to change the default world settings
  • [CLI] adding a print out of running competitors for verbose create_tournament
  • [CLI] adding --world-config to negmas scml
  • [CLI] displaying results of negmas tournament evaluate ordered by the choosen metric in the table.
  • [CLI] preventing very long names
  • [CLI] allowing for more configs/runs in the tournament by not trying all permutations of factory assignments.
  • [CLI] adding --path to negmas tournament create
  • [CLI] more printing in negmas tournament
  • [CLI] reducing default n_retrials to 2
  • [CLI] changing optimism from 0.0 to 0.5
  • [CLI] setting reserved_value to 0.0
  • [CLI] run_tournament does not call evaluate_tournament now
  • [SCML] always adding greedy to std. competitions in negmas tournament
  • [SCML] reducing # colluding agents to 3 by default
  • [CLI] restructuring the tournament command in negmas to allow for pipelining and incremental running of tournaments.
  • [SCML] adding DefaultGreedyManager to manage the behavior of default agents in the final tournament
  • [CLI] avoiding overriding tournament folders if the name is repeated
  • [SCML] avoiding missing reserved_value in some cases in AveragingNegotiatorUfun
  • [CLI] adding the ability to control max-runs interactively to negmas tournament
  • [CLI] adding the ability to use a fraction of all CPUs in tournament with parallel execution
  • [SCML] exceptions in signing contracts are treated as refusal to sign them.
  • [SCML] making contract execution more robust for edge cases (quantity or unit price is zero)
  • [SCML] making collusion tournaments in SCML use the same number of worlds as std tournaments
  • [Situated] adding ignore_contract_execution_excptions to situated and apps.scml
  • [CLI] adding --raise-exceptions/ignore-exceptions to control behavior on agent exception in negmas tournament and negmas scml commands
  • [SCML] adding --path to negmas scml command to add to python path
  • [SCML] supporting ignore_agent_exceptions in situated and apps.scml
  • [Situated] removing total timeout by default

Release 0.2.25

  • [Debugging support] making negmas scml behave similar to negmas tournament worlds
  • [Improved robustness] making insurance calculations robust against rounding errors.
  • [Internal change with no behavioral effect] renaming pay_insurance member of InsuranceCompany to is_insured to better document its nature
  • [Debugging support] adding --balance to negmas scml to control the balance

Release 0.2.24

  • separating PassThroughNegotiator, PassThroughSAONegotiator. This speeds up all simulations at the expense of backward incompatibility for the undocumented Controller pattern. If you are using this pattern, you need to create PassThroughSAONegotiator instead of SAONegotiator. If you are not using Controller or you do not know what that is, you probably safe and your code will just work.
  • adding logging of negotiations and offers (very slow)
  • preventing miners from buying in case sell CFPs are posted.
  • avoiding exceptions if the simulator is used to buy/sell AFTER simulation time
  • adding more stats to the output of negmas scml command
  • revealing competitor_params parameters for anac2019_std/collusion/sabotage. This parameter always existed but was not shown in the method signature (passed as part of kwargs).

Release 0.2.23

  • Avoiding backward incompatibility issue in version 0.2.23 by adding INVALID_UTILITY back to both utilities and apps.scml.common

Release 0.2.22

  • documentation update
  • unifying the INVALID_UTILITY value used by all agents/negotiators to be float("-inf")
  • Added reserved_value parameter to GreedyFactoryManager that allows for control of the reserved value used in all its ufuns.
  • enable mechanism plotting without history and improving plotting visibility
  • shortening negotiator names
  • printing the average number of negotiation rounds in negmas scml command
  • taking care of negotiation timeout possibility in SCML simulations

Release 0.2.21

  • adding avoid_free_sales parameter to NegotiatorUtility to disable checks for zero price contracts
  • adding an optional parameter "partner" to _create_annotation method to create correct contract annotations when response_to_negotiation_request is called
  • Avoiding unnecessary assertion in insurance company evaluate method
  • passing a copy of CFPs to on_new_cfp and on_cfp_removal methods to avoid modifications to them by agents.

Release 0.2.20

  • logging name instead of ID in different debug log messages (CFP publication, rejection to negotiate)
  • bug fix that caused GreedyFactoryManagers to reject valid negotiations

Release 0.2.19

  • logging CFPs
  • defaulting to buying insurance in negmas scml
  • bug resolution related to recently added ability to use LinearUtilityFunction created by a dict with tuple outcomes
  • Adding force_numeric to lead_genius_*

Release 0.2.18

  • minor updates

Release 0.2.17

  • allowing anac2019_world to receive keyword arguments to pass to chain_world
  • bug fix: enabling parameter passing to the mechanism if given implicitly in MechanismFactory()
  • receiving mechanisms explicitly in SCMLWorld and any other parameters of World implicitly

Release 0.2.16

  • bug fix in GreedyFactoryManager to avoid unnecessary negotiation retrials.

Release 0.2.15

  • Minor bug fix to avoid exceptions on consumers with None profile.
  • Small update to the README file.

Release 0.2.14

  • Documentation update
  • simplifying continuous integration workflow (for development)

Release 0.2.13

  • Adding new callbacks to simplify factory manager development in the SCM world
    • on_contract_executed, on_contract_breached
    • on_inventory_change, on_production_success, on_cash_transfer
  • Supporting callbacks including onUfunChanged on jnegmas for SAONegotiator
  • Installing jenegmas 0.2.6 by default in negmas jengmas-setup command

Release 0.2.12

  • updating run scml tutorial
  • tox setting update to avoid a break in latest pip (19.1.0)
  • handling an edge case with both partners committing breaches at the same time.
  • testing reduced max-insurance setting
  • resolving a bug in contract resolution when the same agent commits multiple money breaches on multiple contracts simultaneously.
  • better assertion of correct contract execution
  • resolving a bug in production that caused double counting of some production outputs when multiple lines are executed generating the same product type at the same step.
  • ensuring that the storage reported through awi.state or simulator.storage_* are correct for the current step. That involves a slight change in an undocumented feature of production. In the past produced products were moved to the factory storage BEFORE the beginning of production on the next step. Now it is moved AFTER the END of production of the current step (the step production was completed). This ensures that when the factory manager reads its storage it reflects what it actually have at all times.
  • improving printing of RunningCommandInfo and ProductionReport
  • regenerating setup.py
  • revealing jobs in FactoryState
  • handling a bug that caused factories to have a single line sometimes.
  • revealing the dict jobs in FactoryState which gives the scheduled jobs for each time/line
  • adding always_concede option to NaiveTitForTatNegotiator
  • updating insurance premium percents.
  • adding more tests of NaiveTitForTatNegotiator
  • removing relative_premium/premium confusion. Now evaluate_premium will always return a premium as a fraction of the contract total cost not as the full price of the insurance policy. For a contract of value 30, a premium of 0.1 means 3 money units not 0.1 money units.
  • adding --config option to tournament and scml commands of negmas CLI to allow users to set default parameters in a file or using environment variables
  • unifying the meaning of negative numbers for max_insurance_premium to mean never buying insuance in the scheduler, manager, and app. Now you have to set max_insurance_premium to inf to make the system
  • enforcing argument types in negmas CLI
  • Adding DEFAULT_NEGOTIATOR constant to apps.scml.common to control the default negotiator type used by built-agents
  • making utility_function a property instead of a data member of negotiator
  • adding on_ufun_changed() callback to Negotiator instead of relying on on_nofitication() [relying on on_notification still works].
  • deprecating passing dynamic_ufun to constructors of all negotiators
  • removing special treatment of AspirationNegotiator in miners
  • modifications to the implementation of TitForTatNegotiator to make it more sane.
  • deprecating changing the utility function directly (using negotiator.utility_function = x) AFTER the negotiation starts. It is still possible to change it up to the call to join()
  • adding negmas.apps.scml.DEFAULT_NEGOTIATOR to control the default negotiator used
  • improved parameter settings (for internal parameters not published in the SCML document)
  • speeding up ufun dumping
  • formatting update
  • adding ufun logging as follows:
    • World and SCMLWorld has now log_ufuns_file which if not None gives a file to log the funs into.
    • negmas tournament and scml commands receive a --log-ufuns or --no-log-ufuns to control whether or not to log the ufuns into the tournament/world stats directory under the name ufuns.csv
  • adding a helper add_records to add records into existing csv files.

Release 0.2.11

  • minor bug fix

Release 0.2.10

  • adding more control to negmas tournaments:
    1. adding --factories argument to control how many factories (at least) should exist on each production level
    2. adding --agents argument to control how many agents per competitor to instantiate. For the anac2019std ttype, this will be forced to 1
  • adding sabotage track and anac2019_sabotage to run it
  • updating test assertions for negotiators.
  • tutorial update
  • completed NaiveTitForTatNegotiator implementation

Release 0.2.9

  • resolving a bug in AspirationNegotiator that caused an exception for ufuns with assume_normalized
  • resolving a bug in ASOMechanism that caused agreements only on boundary offers.
  • using jnegmas-0.2.4 instead of jnegmas-0.2.3 in negmas jnegmas-setup command

Release 0.2.8

  • adding commands to FactoryState.
  • Allowing JNegMAS to use GreedyFactoryManager. To do that, the Java factory manager must inherit from GreedyFactoryManager and its class name must end with either GreedyFactoryManager or GFM

Release 0.2.7

  • improving naming of java factory managers in log files.
  • guaranteeing serial tournaments when java factory managers are involved (to be lifter later).
  • adding links to the YouTube playlist in README
  • adhering to Black style

Release 0.2.6

  • documentation update
  • setting default world runs to 100 steps
  • rounding catalog prices and historical costs to money resolution
  • better defaults for negmas tournaments
  • adding warnings when running too many simulations.
  • added version command to negmas
  • corrected the way min_factories_per_level is handled during tournament config creation.
  • added --factories to negmas tournament command to control the minimum number of factories per level.
  • improving naming of managers and factories for debugging purposes
  • forcing reveal-names when giving debug option to any negmas command
  • adding short_type_name to all Entity objects for convenient printing

Release 0.2.5

  • improvements to ufun representation to speedup computation
  • making default factory managers slightly less risky in their behavior in long simulations and more risky in short ones
  • adding jnegmas-setup and genius-setup commands to download and install jenegmas and genius bridge
  • removing the logger mixin and replaced it with parameters to World and SCMLWorld
  • added compact parameter to SCMLWorld, tournament, and world generators to reduce the memory footprint
  • added --compact/--debug to the command line tools to avoid memory and log explosion setting the default to --compact
  • improving implementation of consumer ufun for cases with negative schedule
  • changing the return type of SCMLAWI.state from Factory to FactoryState to avoid modifying the original factory. For efficiency reasons, the profiles list is passed as it is and it is possible to modify it but that is forbidden by the rules of the game.
  • Speeding up and correcting financial report reception.
  • Making bankruptcy reporting system-wide
  • avoiding execution of contracts with negative or no quantity and logging ones with zero unit price.
  • documentation update
  • bug fix to resolve an issue with ufun calculation for consumers in case of over consumption.
  • make the default behavior of negmas command to reveal agent types in their names
  • preventing agents from publishing CFPs with the ID of other agents
  • documentation update
  • improved Java support
  • added option default_dump_extension to ~/negmas/config.json to enable changing the format of dumps from json to yaml. Currently json is the default. This included adding a helper function helpers.dump() to dump in the selected format (or overriding it by providing a file extension).
  • completing compatibility with SCML description (minor change to the consumer profile)
  • added two new options to negmas tournament command: anac2019std and anac2019collusion to simulate these two tracks of the ANAC 2019 SCML. Sabotage version will be added later.
  • added two new functions in apps.scml.utils anac2019_std, anac2019_collusion to simulate these two tracks of the ANAC 2019 SCML. Sabotage version will be added later.
  • added assign_managers() method to SCMLWorld to allow post-init assignment of managers to factories.
  • updating simulator documentation

Release 0.2.2

  • modifications to achieve compatibility with JNegMAS 0.2.0
  • removing the unnecessary ufun property in Negotiator

Release 0.2.0

  • First ANAC 2019 SCML release
  • compatible with JNegMAS 0.2.0

Release 0.1.45

  • implemented money and inventory hiding
  • added sugar methods to SCMLAWI that run execute for different commands: schedule_production, stop_production, schedule_job, hide_inventory, hide_money
  • added a json file ~/negmas/config.json to store all global configs
  • reading jar locations for both jnegmas and genius-bridge from config file
  • completed bankruptcy and liquidation implementation.
  • removed the unnecessary _world parameter from Entity
  • Added parameters to the SCML world to control compensation parameters and default price for products with no catalog prices.
  • Added contract nullification everywhere.
  • updated documentation to show all inherited members of all classes and to show all non-private members
  • Removing the bulletin-board from the public members of the AWI

Release 0.1.42

  • documentation improvement
  • basic bankruptcy implementation
  • bug fixes

Release 0.1.40

  • documentation update
  • implementing bank and insurance company disable/enable switches
  • implementing financial reports
  • implementing checks for bankruptcy in all built-in agents in SCML
  • implementing round timeout in SAOMechanism

Release 0.1.33

  • Moving to Travis CI for continuous integration, ReadTheDocs for documentation and Codacy for code quality

Release 0.1.32

  • Adding partial support to factory manager development using Java
  • Adding annotation control to SCML world simulation disallowing factory managers from sending arbitrary information to co-specifics
  • Removing some unnecessary dependencies
  • Moving development to poetry. Now we do not keep a setup.py file and rely on poetry install

Release 0.1.3

  • removing some unnecessary dependencies that may cause compilation issues

Release 0.1.2

  • First public release

Project details


Release history Release notifications | RSS feed

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

negmas-0.3.2.tar.gz (261.6 kB view hashes)

Uploaded Source

Built Distribution

negmas-0.3.2-py2.py3-none-any.whl (263.0 kB view hashes)

Uploaded Python 2 Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page