Competition Settings and Quick Start Guide

Competition Settings

Rules

Challenge: agents can make use of bids in previous negotiation sessions. Therefore, we explicitly permit the use of local file and a popular machine learning library (scikit-learn).

Platform: Genius Genius 9.1.1
The number of agents: 2
Protocol: Alternative Offer Protocol (AOP)
Negotiation time: 10 sec.
The number of negotiations on the same configuration: 100
Taking over information:
We permit agents to use the default functions (storageexample/GroupX.java including Genius) of loading and reserving the previous negotiation sessions as STANDARD mode. We also permit agents to read/write the local file “your_agent_name.txt” in order to take over the information on past negotiation sessions. The utility of bids, bids’ information, time can take over. We won’t permit taking over the preference information at all. The local file will be removed every time when the opponents or the preferences are changed.
Using Python:
We permit agents to call a Python process.

Parameters of Domains and Profiles

The number of issues: $n=5$
The number of values for each issue: $n_i=5$
Parameter generation:
- $\beta^w \sim Dirichlet$
- $\beta_j^v \sim Beta$
Reservation value: TBD
Discounting factor: TBD

Evaluation

The performance of the agents will be determined by the average individual utilities gained by each agent, and the average social welfares (the sum of individual utilities of each agent) gained by each agent, separately. That means, we have two categories for the winners: individual utility winners and social welfare winners. The teams of the top performing agents will be notified, and the final results and awards will be announced at PRIANAC session. It is expected that teams that make it through to the finals will have a representative attending PRIANAC session. We reserve the right to disqualify agents under certain circumstances. Teams in the final will be given the opportunity to give a brief presentation describing their agent at the session.

Environment(TBD)

OS: CentOS
RAM: 64GB
CPU: 3.6GHz
Platform: Genius 9.1.1
Java: version 1.8.0
Python
- Python 3.6.6
- numpy 1.12.1
- scipy 1.15.1
- scikit-learn 0.19.2
- To use other packages, your agent has to include the source code of the packages.

Quick Start Guide for Automated Negotiation

Negotiation Domain

Domains define negotiation spaces which represent valid bids. In multi-issue negotiations adopted in PRIANAC, domains consist of issues and values of each issue. Agents select a value for each issue and propose it. A set of selected value is called bid.

A simple example is lunch-domain. Issues are Food and Drink, and values are Hamburger, Pizza, and so on. A bid is represented as $(Hamburger, Cola)$ .

Issue	Value 1	Value 2
Food	Hamburger	Pizza
Drink	Cola	Beer

Utility Function

Let $n$ be the number of issues and $\beta_i^w (\sum_{i=0}^n \beta_i^w = 1.0)$ be the weight of each issue $i$ . Utility function $U(\omega)$ of bid $\omega$ is

U (ω) = \sum_{i = 1}^{n} β_{i}^{w} u_{i} (ω_{i})

$\begin{equation*} U(\omega) = \sum_{i=1}^n \beta_i^w u_i( \omega_i) \end{equation*}$
, where

ω_{i} \in {0, 1}^{n_{i}}

$\omega_i \in \{0, 1\}^{n_i}$ is a chosen value in issue

i

$i$ represented as a one-hot vector.
For each issue

i

$i$ , let

n_{i}

$n_i$ be the number of values of the issue and utility function

u_{i} (ω_{i})

$u_i(\omega_i)$ be

u_{i} (ω_{i}) = \sum_{j = 1}^{n_{i}} β_{j}^{v} ω_{i j}

$\begin{equation*} u_i(\omega_i) = \sum_{j=1}^{n_i} \beta_j^v \omega_{ij} \end{equation*}$

, where $\beta_j^v$ is an evaluation value of $j$ and normalized to satisfy $max(\{\beta_j^v \mid 1 \leq j \leq n_i, j \in \mathbb{N} \}) = 1.0$ .

Agents have unique $\beta^w$ and $\beta^v$ to represent its preference. These parameters are defined in a profile.

Now, let us consider lunch-domain. We assume $\beta_i^w$ and $\beta_j^v$ as the following table. Utility function $U(\omega)$ of bid $\omega = (Hamburger, Beer)$ is

U (ω) = β_{F o o d}^{w} β_{H a m b u r g e r}^{v} + β_{D r i n k}^{w} β_{B e e r}^{v} = 0.91.

$\begin{equation} U(\omega) = \beta_{Food}^w \beta_{Hamburger}^v + \beta_{Drink}^w \beta_{Beer}^v = 0.91. \end{equation}$

Issue	Value 1	Value 2
$\beta_{Food}^w = 0.3$	$\beta_{Hamburger}^v = 0.7$	$\beta_{Pizza}^v = 1.0$
$\beta_{Drink}^w = 0.7$	$\beta_{Cola}^v = 0.2$	$\beta_{ Beer}^v = 1.0$

We can explain utility function in vector notation. In lunch-domain, let $\omega = \left( is\_Hamburger, is\_Pizza, is\_Cola, is\_Beer \right)$ be a bid and $\beta = (\beta_{Food}^w \beta_{Hamburger}^v, \beta_{Food}^w \beta_{Pizza}^v, \beta_{Drink}^w \beta_{Cola}^v, \beta_{Drink}^w \beta_{Beer}^v)$ be weights and evaluation values.
Utility function $U(\omega)$ of bid $\omega$ is

U (ω) = β^{T} ω .

$\begin{equation} U(\omega) = \beta^{\mathrm{T}} \omega . \end{equation}$

Hamburger	Pizza	Cola	Beer
$is\_Hamburger = 1$	$is\_Pizza = 0$	$is\_Cola = 0$	$is\_Beer = 1$

Hamburger	Pizza	Cola	Beer
$\beta_{Food}^w \beta_{Hamburger}^v = 0.21$	$\beta_{Food}^w \beta_{Pizza}^v = 0.30$	$\beta_{Drink}^w \beta_{Cola}^v = 0.14$	$\beta_{Drink}^w \beta_{Beer}^v = 0.70$

Negotiation Protocol

The negotiation protocol adopted in PRIANAC is Alternating Offers Protocol (AOP). In AOP, agents take one of the following actions alternately.

Accept: accept opponent’s bid
Offer: reject opponent’s bid and propose new bid
EndNegotiation: end the negotiation without an agreement

Negotiations end in the following cases.

An agent takes Accept.
An agent takes EndNegotiation.
The deadline has passed.

Each agent obtains utility $U(\omega)$ when Accept for $\omega$ is performed. In other cases, each agent obtains a reservation value.

Time Representation

A Negotiation has timeline $t$ , which is normalized to the range $[0; 1]$ . $t = 0$ represents the time of starting a negotiation, and $t = 1$ means the deadline.

Discount Factor

Discount factor decreases agent’s utility depending on $t$ . Discounted utility $U_D(\omega, t)$ is

U_{D} (ω, t) = U (ω) \cdot δ^{t}

$\begin{equation*} U_D(\omega, t) = U(\omega) \cdot \delta^t \end{equation*}$

, where $\delta (0 \leq \delta \leq 1)$ is a discount coefficient define in a profile. $\delta = 1.0$ means utility is independent of $t$ .

Reservation Value

Reservation value is utility of agents when negotiations failed. Reservation value is defined in profiles. Considering discount factor, utility in failed negotiations is

R V_{D} (t) = R V \cdot δ^{t}

$\begin{equation*} RV_D(t) = RV \cdot \delta^t \end{equation*}$

, where $RV (0 \leq RV \leq 1)$ is a reservation value define in a profile.

Useful Links

Following pages would help you when you make an agent.