POLIXIR REVIVE For VS Code

About

POLIXIR REVIVE is a development tool based on VS Code for developers to use REVIVE SDK more conveniently and quickly.
It provides a series of functions such as sdk installation,decision flow design,hyperparameter management,nteractive model inference,log analysist,etc.
In addition,it also provides detailed description of teaching cases and help documents for easy introduction.
More about POLIXIR REVIVE SDK please visit https://revive.cn/sdk.

Installation

POLIXIR REVIVE VSCode Extension provides installation instructions and one-click operations for sdk.
To install the extension,open the Extensions view,search and select POLIXIR REVIVE extension authored by POLIXIR.

Create a new REVIVE SDK development

If you are using REVIVE SDK for the first time, or want to create a new development project with the latest SDK,you can use the extension to install REVIVE SDK by following the instructions.

Open an existing folder of REVIVE SDK development

If you have installed REVIVE SDK,you only need to open the installed sdk directory to start your work.

Account

Register in revive

To use REVIVE SDK for development,you first need to pass online authentication. If you are using it for the first time and have not applied for an account,click "Register in revive" and register your own account.

Authentication

If you already have an account,just need to authenticate by click authentication.We support three authentication methods: email,mobile number and username.

Overview of the extension's toolkit

Detect the running environment of the SDK

For many developers,the development environment is the first threshold.
POLIXIR REVIVE For VS Code Extension provides a detection tool for the development environment.
If you are not sure about your development environment,execute it and follow the instructions to install some dependencies.

Quick Start

POLIXIR REVIVE extension provide a quick start feature,you just need to select the relevant configuration file and code file,and you can start training.

Decision Flow

Draw a simple decision flow

Nodes of either States,Agent,Environment Agent,Static Variables can be dragged to the drawing area to build up the decision flow.

• States represents variables of the environment states.Even state node is automatically attached with a State-transition node and another state node.The two state nodes indicates the variables in the current step and in the next step.The State-transition node aggregate inputs and outputs the states in the next step.
• Agent represents the agent model that will be trained in the reinforcement learning stage for maximizing the total reward.
• Environment Agent represents a model in the environment.It is part of the environment after the environment leanring stage,and will be fixed in the reinforcement learning stage.
• Static Variables represents variables that are not effected by the modeled system,such as the weather.

To express the affecting relationships between nodes,you can drag the arrow of one node to the head of another node to show a directed edge.The edge represents the causal effect between the two nodes. (An illustration of the decision-flow)

Each node needs to be associated with data dimensions to complete its field settings.

First, you need to import data:

Then you can do field setting and field assignment.

Add Multi-Time-Steps Data Nodes

To improve model accuracy, it is often possible to use historical multi-time-steps data as input. This extension plug-in provides an easy way to configure multi-time-step nodes.

Append time step index as node input

The time node is the time step of the data. By configuring the time node information, you can use the time step index information as the input of the node.

Add expert function to the node

By default,a node is implemented as a neural network.However,you may want to have a hand-crafted function as a node or a manually set neural network. Expert function can realize this function

Convert data to npz or h5

The extension plug-in provides a flexible data conversion function, which can convert the csv data into npz or h5 by combining it with the decision flow.

Causal Analysis

By analyzing the causal relationship between data fields, the auxiliary drawing of decision flow diagram

Conversion between decision flow and yaml file

The decision flow .prdf.json files and .yaml files can be converted to each other.You can press the button SAVE AS YAML or choose a decison flow, right click and choose Save as Yaml.

Manage hyperparameter

POLIXIR REVIVE extension support two kinds of hyperparameter :enviroment and policy.

• Environment hyperparametercontains base hyperparameter,BC hyperparameter and revive hyperparameter.
• Policy hyperparameter contains base hyperparameter,SAC hyperparameter and PPO hyperparameter.

POLIXIR REVIVE extension support Visual interface to help developer build hyperparameter file quickly.

Performance Board

POLIXIR revive extension provides management functions for training logs,after selecting the log path,the log will be displayed in a visual way,including key indicators,charts,etc. It also support open tensorboard for analysis.

Interactive Model Inference

After training the environment and policy, the trajectory of environment interaction can be deduced..
when the initial data is determined, the function will default to deduce a 10 step path backwards to show the status and action content of each step.
You can change the state or action content in any step to deduce the changed track.

Teaching cases

The extension plug-in provides teaching cases, including descriptions, data and executable code, for developers to quickly get started.

Help and feedback

In order to encourage more developers to use REVIVE SDK,we have provided a complete help center and teaching videos,and established a forum discussion area for REVIVE SDK,which is convenient for everyone to communicate with each other.