Deploying Trains Overview
Trains is now ClearML
This documentation applies to the legacy Trains versions. For the latest documentation, see ClearML.
This page describes the Trains open source suite deployment options, which include:
- The demo Trains Server, which works with Trains by default, is open to anyone, and resets periodically deleting data.
- A self-hosted Trains Server
- Trains Agent (our DevOps solution for experiment execution and automation)
The Trains Python Package
To install and optionally set configuration options:
-
Install the Trains package.
pip install trains
Now, just add two lines of code to your Python experiment script.
from trains import Task task = Task.init(project_name="my project", task_name="my task")
When you run your script, Trains automagically monitors and logs its results which you can track and compare in the demo Trains Server (https://demoapp.trains.allegro.ai/dashboard) Trains Web-App (UI).
-
Optionally, configure Trains options (for example, cache, metrics, and log settings), see the Trains Configuration Reference.
Trains Server and Trains Agent deployment options
Option 1. The demo Trains Server, only.
Automagically track, analyze, and compare experiments using the demo Trains Web-App (UI).
Our demo Trains Server is periodically refreshed and data is deleted.
The Trains Python Package automatically works with the demo Trains Server.
Optionally, configure Trains options (for example, cache, metrics, and log settings), see the Trains Configuration Reference.
Option 2. The demo Trains Server and Trains Agent
Manage your remote experiment execution using Trains Agent (see Task remote execution), but not your own backend infrastructure.
To install and optionally set configuration options:
- Install and Configure Trains Agent.
- Optionally, configure Trains options (for example, cache, metrics, and log settings), see the Trains Configuration Reference.
Option 3. A self-hosted Trains Server
Manage your own Trains backend infrastructure, including your own Trains Web-App (UI).
Overview of deployment:
-
Deploy your own Trains Server, using one of the available formats which include:
- Pre-built AWS EC2 AMIs.
- Pre-built Google Cloud Platform AMIs.
- Pre-built Docker images for Linux, macOS, and Windows 10.
- Kubernetes and Kubernetes using Helm.
-
Optionally, configure Trains Server for additional features, including sub-domains and load balancers, web login authentication, and the non-responsive task watchdog.
Option 4. A self-hosted Trains Server and Trains Agent.
Manage your own Trains backend infrastructure, and manage your remote experiment execution using Trains Agent.
Overview of deployment:
- Perform the steps in Option 3.
- Install and Configure Trains Agent.