river-core
v0.1.0
Published
Library for building computational graphs
Downloads
2
Maintainers
Readme
River Core
"River Core" is a minimal set of functions for building computational graphs.
Computational graphs, in essence, are networks of functions. But unlike common functions, River nodes receive their inputs and emit their outputs independently from one another, and instead of parameters, values get passed in and emitted through ports.
Getting started
npm i river-core
API
The River core API consists of three functions:
createNode(outFields, createInPorts)
connect(outPort, inPort)
disconnect(outPort[, inPort])
createNode(outFields, createInPorts)
Creates a node with the specified output and input ports. As output ports have no logic associated with them, only their names are required.
The function createInPorts()
takes one argument, outputs
' which is a
lookup of functions emitting on the node's output ports, and returns
functions in a similar structure but for input ports.
Both input and output functions take two arguments:
value
, which is the value being sent ot / emitted by the node, andtag
, which identifies the impulse throughout the graph.
createNode()
returns a Node
object, with at least two properties:
i
, a collection of functions that serve as the node's input portso
, a collection of structures that serve as output ports (the specifics of these structures are not relevant to their usage)
Example / TypeScript
import {createNode} from "river-core";
type In = {d_in: number};
type Out = {d_out: number};
const node = createNode<In, Out>(["d_out"], (outputs) => ({
d_in: (value, tag) => outputs.d_out(value, tag)
}));
Example / ES6
const core = require("river-core");
const node = core.createNode(["d_out"], (outputs) => ({
d_in: (value, tag) => outputs.d_out(value, tag)
}));
connect(outPort, inPort)
Connects an output port to an input port. After a connection is made, when a node emits a value on one of its connected output ports, that value will be used to invoke all connected input ports.
It's worth noting that input ports are simply functions, and therefore it's
possible to 'connect' an output port to a function, eg. console.log
.
Example / TypeScript
import {connect} from "river-core";
// initializing node1 & node2
connect(node1.o.d_out, node2.i.d_in);
Example / ES6
const core = require("river-core");
// initializing node1 & node2
core.connect(node1.o.d_out, node2.i.d_in);
disconnect(outPort[, inPort])
Disconnects an input port from an output port, or, disconnect all input ports from an output port.
Example / TypeScript
import {disconnect} from "river-core";
// initializing & connecting node1 & node2
disconnect(node1.o.d_out, node2.i.d_in);
disconnect(node1.o.d_out);
Example / ES6
const core = require("river-core");
// initializing & connecting node1 & node2
core.disconnect(node1.o.d_out, node2.i.d_in);
core.disconnect(node1.o.d_out);
Types
River Core exports the following types:
Any
: Lookup with string keysInPort
: Describes an input portInPorts
: Describes a set of input ports. Usually corresponds to thei
property of nodes.Node<In, Out>
: Describes a node, whereIn
andOut
describe the value type carried by each input and output port, respectively.OutPort
: Describes an output portOutPorts
: Describes a set of input ports. Usually corresponds to theo
property of nodes.Outputs
: Describes a set of functions that invoke connected input ports.Tag
: Identifies impulses. (In turn, an impulse is an identifiable input that ripples through multiple nodes throughout the graph.)
Custom nodes
In practice, createNode()
is not invoked directly, as one would have to
supply types (for TypeScript), output port names, and the function
createInPorts()
on each call. Instead, createNode()
usually gets wrapped
inside an outer factory function which has these already bundled.
The example below will produce the same node as the ones above, in a reusable way. For those coming from an OOP background, this would be analogous to subclassing.
Example / TypeScript
import {createNode, Node} from "river-core";
type In = {d_in: number};
type Out = {d_out: number};
type Forwarder<V> = Node<In, Out>
function createForwarder<V>(): Node {
return createNode<In, Out>(["d_out"], (outputs) => ({
d_in: (value, tag) => outputs.d_out(value, tag)
}));
}
Example / ES6
const core = require("river-core");
function createForwarder() {
return createNode(["d_out"], (outputs) => ({
d_in: (value, tag) => outputs.d_out(value, tag)
}));
}
Composite nodes
One of the most useful features of computational graphs is their composability. A number of interconnected nodes may act as as a single node themselves, making it easier to compartmentalize functionality, as well as reason about different parts of the application.
Implementing a composite node is quite simple with a factory function, which will create multiple nodes, establish internal connections, and expose specific ports of its components as its own.
Example / TypeScript
import {connect, Node} from "river-core";
type In = {d_in: number};
type Out = {d_out: number};
type Composite = Node<In, Out>;
function createComposite<In, Out>(): Composite {
const forwarder1 = createForwarder();
const forwarder2 = createForwarder();
connect(forwarder1.o.d_out, forwarder2.i.d_in);
return {
i: {d_in: forwarder1.i.d_in},
o: {d_out: forwarder2.o.d_out}
};
}
Example / ES6
const core = require("river-core");
function createComposite() {
const forwarder1 = createForwarder();
const forwarder2 = createForwarder();
connect(forwarder1.o.d_out, forwarder2.i.d_in);
return {
i: {d_in: forwarder1.i.d_in},
o: {d_out: forwarder2.o.d_out}
};
}