Skip to content

Getting started#

In this tutorial, you will learn how to define and execute experiments, trying the key features. You can follow it sequentially or jump to specific questions as needed.

Installation#

MLtraq requires Python 3.10+ and depends on SQLAlchemy 2.0+, Pandas 1.5.3+, and Joblib 1.3.2+, which are installed as dependencies. To install:

pip install mltraq --upgrade

Examples#

The code examples are fully self-contained to reproduce the outputs. In this example, the version of MLtraq used to compile this tutorial is shown. Make sure to have the latest release installed.

MLtraq version

1
2
3
import mltraq

print(mltraq.__version__)
Output
0.1.136

Key concepts#

  • Experimentation: The process of systematically changing and testing different input values in an algorithm to observe their impact on performance, behavior, or outcomes. Experiments can be defined and executed, with their outcomes and/or results persisted for later analysis.

  • Session: A session object lets you define the connection to a database, load and add experiments. Sessions are bound to a database.

  • Experiment: An experiment object manages a collection of run objects. Experiments can be created, persisted, loaded and executed. It implements the experimentation process. A run is an instantiation of the experiment with a configuration of input values. The execution of an experiment requires the execution of all its runs. Experiments are bound to a database and are unaware of sessions.

  • Run: A run object is an instantiation of the experiment with a configuration of input values. The execution of a run is defined as the chained evaluation of step functions, whose sole parameter is the run object itself. Runs are unaware of databases, sessions, experiments or other runs, and are isolated from the rest of the experiment.

  • Step: Step functions are a Python functions that take as sole input the run object, changing its internal state. There is no return value. Steps can access the configuration of the run in the attributes run.config and run.params, and can change the state of the run by modifying the attributes run.vars, run.state and run.fields.

Tip

An overview of the run state attributes can be found in State management, with a discussion of their semantics in the Model of computation.