README.md

RLPerf: Benchmark for Autonomous Agents.

RLPerf is a benchmark for evaluating agents on sequential decision problems that are relevant to the real world. This repository contains code for running and evaluating participant's submissions on the benchmark platform.

Installation

RLPerf can be installed on your local machine:

git clone https://github.com/harvard-edge/rl-perf.git
cd rl-perf
git submodule init --update --recursive
pip install -e .

---

Domains

RLPerf provides benchmark environments in the following domains:

Web Navigation

Quadruped Locomotion

Chip Floorplanning

Gym Interface

Environments in RLPerf are registered under the names WebNavigation-v0, QuadrupedLocomotion-v0, and CircuitTraining-v0. For example, you can create an instance of the WebNavigation-v0 environment as follows:

from rl_perf.domains.web_nav import web_nav
import gym

env = gym.make('WebNavigation-v0', difficulty=1, seed=0)

---

Participant Submission

• Participants can pull the template repository at https://github.com/harvard-edge/rlperf_benchmark_submission
  • The submission repository must include:
    • train.py - defines a train function that can train the participant's model from scratch
    • train.gin - defines configurations for how the participant wants their solution to be benchmarked. For example, thhe participant could include values for these contants

@gin.constants_from_enum
class BenchmarkMode(enum.Enum):
    TRAIN = 'train'
    EVAL = 'eval'
    INFERENCE = 'inference'


@gin.constants_from_enum
class BenchmarkDomain(enum.Enum):
    WEB_NAVIGATION = 'web_navigation'
    CIRCUIT_TRAINING = 'circuit_training'
    QUADRUPED_LOCOMOTION = 'quadruped_locomotion'


@gin.constants_from_enum
class ReliabilityMetrics(enum.Enum):
    IqrWithinRuns = 'IqrWithinRuns'
    IqrAcrossRuns = 'IqrAcrossRuns'
    LowerCVaROnDiffs = 'LowerCVaROnDiffs'
    LowerCVaROnDrawdown = 'LowerCVaROnDrawdown'
    LowerCVarOnAcross = 'LowerCVarOnAcross'
    MedianPerfDuringTraining = 'MedianPerfDuringTraining'

---

Baselines

TODO

Credits

This package is maintained by Ikechukwu Uchendu and Jason Jabbour.