Overview

django_analyses provides a database-supported pipeline engine meant to facilitate research management.

A general schema for pipeline management is laid out as follows:

Analyses

Each Analysis may be associated with a number of AnalysisVersion instances, and each of those must be provided with an interface, i.e. a Python class exposing some run() method and returning a dictionary of results.

For more information, see the Simplified Analysis Integration Example.

Input and Output Specifications

InputSpecification and OutputSpecification simply aggregate a number of InputDefinition and OutputDefinition sub-classes (respectively) associated with some analysis.

Input and Output Definitions

Currently, there are seven different types of built-in input definitions:

• BooleanInputDefinition
• DirectoryInputDefinition
• FileInputDefinition
• FloatInputDefinition
• IntegerInputDefinition
• ListInputDefinition
• StringInputDefinition

and two different kinds of supported output definitions:

• FileOutputDefinition
• FloatOutputDefinition

Each one of these InputDefinition and OutputDefinition sub-classes provides unique validation rules (default, minimal/maximal value or length, choices, etc.), and you can easily create more definitions to suit your own needs.

Pipelines

Pipeline instances are used to reference a particular collection of Node and Pipe instances.

• A Node is defined by specifying a distinct combination of an AnalysisVersion instance and a configuration for it.
• A Pipe connects between a one node’s output definition and another’s input definition.

For more information, see Pipeline Generation.

Runs

Run instances are used to keep a record of every time an analysis version is run with a distinct set of inputs, and associate that event with the resulting outputs.

Whenever a node is executed, the value assigned to each of the InputDefinition model’s sub-classes detailed in that interface’s InputSpecification is committed to the database as the corresponding Input model’s sub-class instance.

If we ever execute a run with identical parameters, the RunManager will simply return the existing run.