Skip to content

Latest commit

 

History

History
executable file
·
184 lines (117 loc) · 4.54 KB

README.md

File metadata and controls

executable file
·
184 lines (117 loc) · 4.54 KB

Skynet

Skynet is a service designed for advanced model management, transformation, ranking, and visualization. It supports dynamic APIs for working with machine learning models and provides extensive health checks, metrics, and documentation.

Features

  • Model Loading & Management: Load, unload, and list models dynamically via API.
  • Data Transformation: Transform data using preloaded models.
  • Ranking Framework: Rank model outputs with customizable parameters.
  • Visualization: Generate computation graphs of loaded models.
  • Health Checks: Monitor both service and model health.
  • Swagger Documentation: Auto-generated OpenAPI specs with Swagger UI.
  • Metrics: Comprehensive monitoring and metrics for endpoints.

Getting Started

Prerequisites

  • Scala 2.13. and sbt for building and running the service.
  • Docker for containerization (optional).
  • Dependencies are managed in the SBT build file.

Application Structure

  • Main Entry Point: Starter.scala

    • Bootstraps the service, initializes configurations, and starts the server.
    • Uses ZIO for dependency injection and environment management.

  • API Definition: Endpoints.scala

    • Defines RESTful API endpoints using Tapir.
    • Includes model management, transformation, and health check APIs.

  • Routing: Routes.scala

    • Maps endpoints to route handlers and integrates Swagger for API documentation.
    • Includes middleware for metrics and error handling.

  • Build Configuration: build.sbt

    • Handles project dependencies, build plugins, and Docker configurations.

Key API Endpoints

Model Operations

  • Load a Model
    PUT /models/{model}
    Loads a model from a given URI (e.g., file://, s3://).

  • Unload a Model
    DELETE /models/{model}
    Unloads a model from the service.

  • List All Models
    GET /models
    Lists all currently loaded models.

  • Model Health Check
    GET /models/{model}/health
    Performs a test prediction to verify the model is operational.

Data Operations

  • Transform Data
    POST /models/{model}/transform
    Transforms input data using the specified model.

  • Rank Data
    POST /models/{model}/rank
    Ranks transformed data with options for grouping and averaging.

  • Get Sample Data
    GET /models/{model}/sample
    Returns a sample data frame for a model.

Visualization

  • Model Graph
    GET /models/{model}/graph
    Visualizes the computation graph of the model.

Service Health

  • Service Health Check
    GET /health
    Confirms the service is operational.

Build & Run Instructions

Build

sbt compile

Run

sbt run

The service will start and provide API access via http://localhost:8080.

Docker Build

A Docker image can be built using:

sbt docker:publishLocal

This will create a containerized version of the service.

Dependencies

Core Libraries

  • ZIO: Functional effect handling.
  • Tapir: API definition and OpenAPI documentation.
  • Http4s: Web server and client.

Machine Learning

  • MLeap: Runtime and executor for machine learning pipelines.
  • XGBoost: Predictor integration. The implementation here is much faster than MLeaps original implementation. This allows for BLAZINGLY FAST XGBoost serving!

Configuration

  • PureConfig: Simplified configuration management.
  • Typesafe Config: Configuration library.

JSON & Serialization

  • Circe: JSON serialization and parsing.
  • Protobuf: Support for protocol buffers.

Logging

  • Logback: Logging framework.
  • Scala Logging: Integration for logging in Scala.

Documentation

  • Swagger UI: Interactive API documentation.
  • RefTree: Visualization of computation graphs.

Metrics

  • Prometheus: Metrics collection and integration.

API Documentation

Swagger UI is available at: http://localhost:8080/swagger-ui

This provides an interactive interface to explore and test the API.

Metrics

Metrics are exposed at: http://localhost:8080/metrics

These include endpoint-specific metrics and overall service health data.

Contributing

Feel free to contribute by submitting issues or pull requests. Follow functional programming principles and the existing code style.