AIOpsArena SDK User Guide

Introduction

👏 Welcome to the AIOpsArena SDK, a platform's Algorithm Hot-plugging feature designed for deployment model. This platform is dedicated to model deployment services. With its help, users can easily train models and test it. During operation, the platform automatically creates a Docker container to run the algorithm. If you choose to deploy the algorithm online, you can use the train method to train your model and then use the test method to test it. Alternatively, if you just want to observe the effects of the model, you can choose to use the run method to train, test, and obtain results in one go. The path to the original data is written in the system configuration file, which can be directly used in the model methods.

Getting Started

Cloning the repository

First, clone our repository by running the following command:

git clone https://github.com/AIOpsArena/hot-plugging.git

Adding your algorithm code

After cloning the repository, place your own algorithm in the algorithm folder.

Reformat your algorithm code

Before you start implementing the training and testing methods for your algorithm, you need to make some modifications to your code to meet our requirements. Initially, all file path information must be integrated into a configuration file at the algorithm directory, using absolute paths. Adherence to naming conventions is mandatory; directories should end with 'dir' and files with 'path'. Subsequently, users must prepare a requirements.txt file detailing all necessary dependencies for the algorithm’s execution.

Extending the Algorithm Class

After you have adjusted the format, you can start to extend the Algorithm Class. This class is located in algorithmTemplate/algorithm_app/algorithm.py. If you simply want to view the effects of the model without saving it, you only need to implement the __init__ and run methods.

If you want to deploy the algorithm online to the platform, you will need to implement the init, train, and test methods. If there are intermediate results that need to be saved, the corresponding fields in the configuration file should be placed in the 'ignore_list' in 'algorithm_app/public_function.py'.

Before starting, we need to import our own algorithm methods into algorithm.py. We recommend that you encapsulate your algorithm into a class similar to our Algorithm class.

Implementing the __init__ Method

The __init__ method is responsible for handling the path information of the algorithm, initializing your model, and providing some variables for the original data. After processing by the __init__ method, all paths will be modified to start with the algorithmTemplate/algorithm_app/experiment prefix, facilitating subsequent management.

Implementing the train Method

The train method is used to train your model. AIOpsArena will provide the original data (with labels) mentioned in the __init__ method. Here, you need to implement your data preprocessing logic, then use the preprocessed data to train the model, and save the model in the experiment path after training is complete.

Implementing the test Method

The test method is used to test the model. AIOpsArena will read the model trained by the train method and the test data (with labels). In this method, you need to return Json data in a specific format. AIOpsArena will store the data in the database and display it on the platform's evaluation data page.

Implementing the run Method

The run method does not differentiate between training and testing. AIOpsArena provides the original data (with labels). In this method, you need to implement data preprocessing, model training, model testing, and then return Json data in a specific format. Finally, AIOpsArena will store the data in the database.

Example

We have provided two example files, one with split train-test split_example and one without split train-test nosplit_example, hoping to be of some help to you.

Feel free to adapt this example to your specific needs and customize your algorithms for AIOpsArena platform. 🎉🎉

Test your algorithm

After you have completed all operations, don't rush to upload; you can first test it locally. 😎

1. First, build docker image. Run the following command:

   docker build -t YOUR_IMAGE_NAME .

2. Second, run algorithm container.

   docker run --name 'YOUR_CONTAINER_NAME' -v $(pwd):/code YOUR_IMAGE_NAME python manage.py runserver 0.0.0.0:8000

3. Third, inspect container's ip

   docker inspect -f '{{range .NetworkSettings.Networks}}{{.IPAddress}}{{end}}' YOUR_CONTAINER_NAME

4. Fourth, implement the interface of algorithm_app/views.py:test. Such as algorithm.train() and algorithm.test().

5. Fifth, copy the demo dataset to algorithm_app/experiment.

   Create a USER folder under algorithm_app/experiment, and place the demo dataset in USER/origin_data

6. Sixth, send the curl request.

   curl "http:/CONTAINER_IP:8000/test?user=USER"