XAIport

Overview

XAIport is designed to deliver interpretable AI model predictions through a microservice architecture, allowing users to understand the underlying decision-making processes of AI models better. The architecture includes a User Interface, Coordination Center, Core Microservices such as Data Processing, AI Model, XAI Method, and Evaluation Services, along with a Data Persistence layer.

Initial Setup

Prerequisites

• Python 3.8 or later
• FastAPI
• httpx
• uvicorn
• Dependencies as listed in requirements.txt

Installation Guide

1. Environment Setup: Ensure Python is installed on your system. It's recommended to use a virtual environment for Python projects:

   python -m venv venv
   source venv/bin/activate  # On Windows use `venv\Scripts\activate`

2. Install Dependencies: Install the necessary Python libraries with pip:

   pip install -r requirements.txt

3. Clone the Repository: Clone the repository to get the latest codebase:

   git clone https://github.com/ZeruiW/XAIport.git
   cd XAIport

Configuration

Before running the system, configure all necessary details such as API endpoints, database connections, and other service-related configurations in a JSON format. Adjust the config.json file as needed.

Example config.json:

{
  "upload_config": {
    "server_url": "http://localhost:8000",
    "datasets": {
      "dataset1": {
        "local_zip_path": "/path/to/dataset1.zip"
      }
    }
  },
  "model_config": {
    "base_url": "http://model-service-url",
    "models": {
      "model1": {
        "model_name": "ResNet50",
        "perturbation_type": "noise",
        "severity": 2
      }
    }
  }
}

Running the System

Starting the Service

Run the FastAPI application using Uvicorn as an ASGI server with the following command:

uvicorn main:app --host 0.0.0.0 --port 8000

Using the API

The system provides several RESTful APIs to support operations such as data upload, model prediction, XAI method execution, and evaluation tasks. Here are some examples of how to use these APIs:

• Upload Dataset:

  curl -X POST "http://localhost:8000/upload-dataset/dataset1" -F "file=@/path/to/dataset.zip"

• Execute XAI Task:

  curl -X POST "http://localhost:8000/cam_xai" -H "Content-Type: application/json" -d '{"dataset_id": "dataset1", "algorithms": ["GradCAM", "SmoothGrad"]}'

Maintenance and Monitoring

Logging

Configure appropriate logging policies to record key operations and errors within the system. This can be achieved by setting up Python's logging module to handle different log levels and outputs.

Performance Monitoring

It is recommended to use monitoring tools like Prometheus and Grafana to track system performance and health indicators.

Frequently Asked Questions (FAQ)

How do I handle data upload failures?

Check if the target server is reachable and ensure that the file paths in the configuration file are correctly specified.

How do I update API endpoints in the configuration file?

Modify the API endpoints directly in the JSON configuration file and restart the service to apply changes.