CRayGraph

A toolkit for defining dataflow-based frameworks. Intended for use in CRayNN (network definitions) and CRayFlow (dataflow definitions).

Installation

via PyPi

pip install craygraph

via git

CRayGraph can be installed directly from gitlab.com:

pip install git+https://gitlab.com/craynn/craygraph.git

however, as repository updates quite often, it is recommend to clone the repository and install the package in development mode:

git clone git@gitlab.com:craynn/craygraph.git
cd craygraph/
pip install -e .

Usage

This package provides three main functions:

• cragraph.graph : utilities for constructing easily readable definitions of directed acyclic graphs (DAGs);
• cragraph.meta : functions that help quickly adopt custom classes to work with cragraph.graph.

DAG definition language

The main feature of craygraph is intuitive and readable DAG definition language.

Quick introduction

Below, functions like a, b, c, d represent node (subgraph) constructors, i.e. (*Node) -> Node or (*Node) -> list[Nodes], x, y, z represent input nodes.

Tuple of functions = composition

achain(a, b, c, d)(x)

is equivalent to:

d(c(b(a(x))))

and produces:

List of functions = independent application:

result = achain(a, [b, c], d)(x)

is equivalent to:

a_x = a(x)
result = d(b(a_x), c(a_x))

and results in the following graph:

craygraph.graph contains some helper functions. select[items](body) selects inputs according to items and applies body

select[item](f)(*args) == f(args[item]) 

for example:

result = achain(
  select[0, -1](a, b),
  c
)(x, y, z)

is equivalent to:

result = c(b(a(x, z)))

which results in (node y is omitted since it is not used):

with_inputs[items](body) is similar to select, but it takes inputs and replaces selected inputs with the result of body:

result = achain(
  with_inputs[0, -1](a, b),
  c
)(x, y, z)

is equivalent to:

result = c(b(a(x, z)))

which results in: