QTRAN

Overview

QTRAN is proposed by Kyunghwan et al.(2019). QTRAN is a factorization method for MARL, which is free from such structural constraints and takes on a new approach to transform the original joint action-value function into an easily factorizable one, with the same optimal actions.

Compared to VDN(Sunehag et al. 2017), QMIX(Rashid et al. 2018), QTRAN guarantees more general factorization than VDN or QMIX, thus covering a much wider class of MARL tasks than does previous methods, and it performs better than QMIX in 5m_vs_6m and MMM2 maps.

Quick Facts

1. QTRAN uses the paradigm of centralized training with decentralized execution.

2. QTRAN is a model-free and value-based method.

3. QTRAN only support discrete action spaces.

4. QTRAN is an off-policy multi-agent RL algorithm.

5. QTRAN considers a partially observable scenario in which each agent only obtains individual observations.

6. QTRAN accepts DRQN as individual value network.

7. QTRAN learns the joint value function through Individual action-value network, Joint action-value network and State-value network.

Key Equations or Key Graphs

The overall QTRAN architecture including individual agent networks and the mixing network structure:

QTRAN trains the mixing network via minimizing the following loss:

\[L_{\mathrm{td}}(; \boldsymbol{\theta}) =\left(Q_{\mathrm{jt}}(\boldsymbol{\tau}, \boldsymbol{u})-y^{\mathrm{dqn}}\left(r, \boldsymbol{\tau}^{\prime} ; \boldsymbol{\theta}^{-}\right)\right)^{2}\]

\[L_{\mathrm{opt}}(; \boldsymbol{\theta}) =\left(Q_{\mathrm{jt}}^{\prime}(\boldsymbol{\tau}, \overline{\boldsymbol{u}})-\hat{Q}_{\mathrm{jt}}(\boldsymbol{\tau}, \overline{\boldsymbol{u}})+V_{\mathrm{jt}}(\boldsymbol{\tau})\right)^{2}\]

\[L_{\mathrm{nopt}}(; \boldsymbol{\theta}) =\left(\min \left[Q_{\mathrm{jt}}^{\prime}(\boldsymbol{\tau}, \boldsymbol{u})-\hat{Q}_{\mathrm{jt}}(\boldsymbol{\tau}, \boldsymbol{u})+V_{\mathrm{jt}}(\boldsymbol{\tau}), 0\right]\right)^{2}\]

\[L\left(\boldsymbol{\tau}, \boldsymbol{u}, r, \boldsymbol{\tau}^{\prime} ; \boldsymbol{\theta}\right)=L_{\mathrm{td}}+\lambda_{\mathrm{opt}} L_{\mathrm{opt}}+\lambda_{\mathrm{nopt}} L_{\mathrm{nopt}},\]

Pseudo-code

The following flow charts show how QTRAN trains.

Extensions

• QTRAN++ (Son et al. 2019), as an extension of QTRAN, successfully bridges the gap between empirical performance and theoretical guarantee, and newly achieves state-of-the-art performance in the SMAC environment.

Implementations

The default config is defined as follows:

class ding.policy.qtran.QTRANPolicy(cfg: easydict.EasyDict, model: Optional[torch.nn.modules.module.Module] = None, enable_field: Optional[List[str]] = None)

Overview:

Policy class of QTRAN algorithm. QTRAN is a multi model reinforcement learning algorithm, you can view the paper in the following link https://arxiv.org/abs/1803.11485

Config:

ID	Symbol	Type	Default Value	Description	Other(Shape)
1	type	str	qtran	RL policy register name, refer to registry POLICY_REGISTRY	this arg is optional, a placeholder
2	cuda	bool	True	Whether to use cuda for network	this arg can be diff- erent from modes
3	on_policy	bool	False	Whether the RL algorithm is on-policy or off-policy
	priority	bool	False	Whether use priority(PER)	priority sample, update priority
5	priority_ IS_weight	bool	False	Whether use Importance Sampling Weight to correct biased update.	IS weight
6	learn.update_ per_collect	int	20	How many updates(iterations) to train after collector’s one collection. Only valid in serial training	this args can be vary from envs. Bigger val means more off-policy
7	learn.target_ update_theta	float	0.001	Target network update momentum parameter.	between[0,1]
8	learn.discount _factor	float	0.99	Reward’s future discount factor, aka. gamma	may be 1 when sparse reward env

The network interface QTRAN used is defined as follows:

ding.model.template.qtran

alias of <module ‘ding.model.template.qtran’ from ‘/home/docs/checkouts/readthedocs.org/user_builds/di-engine-docs/envs/latest/lib/python3.9/site-packages/ding/model/template/qtran.py’>

The Benchmark result of QTRAN in SMAC (Samvelyan et al. 2019), for StarCraft micromanagement problems, implemented in DI-engine is shown.

smac map	best mean reward	evaluation results	config link	comparison
MMM	1.00		config_link_0	Pymarl(1.0)
3s5z	0.95		config_link_1	Pymarl(0.1)
5m6m	0.55		config_link_2	Pymarl(0.7)

References

• QTRAN: Learning to Factorize with Transformation for Cooperative Multi-Agent Reinforcement Learning. ICML, 2019.

Other Public Implementations

• Pymarl.