36 lines
1.2 KiB
Markdown
36 lines
1.2 KiB
Markdown
|
description: How to use the Random model network in Neataptic
|
||
|
|
authors: Thomas Wagenaar
|
||
|
|
keywords: recurrent, feed-forward, gates, neural-network, random, architecture
|
||
|
|
|
||
|
|
A random network is similar to a liquid network. This network will start of with a given pool of nodes, and will then create random connections between them. This network is really only useful for the initialization of the population for a genetic algorithm.
|
||
|
|
|
||
|
|
```javascript
|
||
|
|
new architect.Random(input_size, hidden_size, output_size, options);
|
||
|
|
```
|
||
|
|
|
||
|
|
* `input_size` : amount of input nodes
|
||
|
|
* `hidden_size` : amount of nodes inbetween input and output
|
||
|
|
* `output_size` : amount of output nodes
|
||
|
|
|
||
|
|
Options:
|
||
|
|
* `connections` : amount of connections (default is `2 * hidden_size`, should always be bigger than `hidden_size`!)
|
||
|
|
* `backconnections` : amount of recurrent connections (default is `0`)
|
||
|
|
* `selfconnections` : amount of selfconnections (default is `0`)
|
||
|
|
* `gates` : amount of gates (default is `0`)
|
||
|
|
|
||
|
|
For example:
|
||
|
|
|
||
|
|
```javascript
|
||
|
|
var network = architect.Random(1, 20, 2, {
|
||
|
|
connections: 40,
|
||
|
|
gates: 4,
|
||
|
|
selfconnections: 4
|
||
|
|
});
|
||
|
|
|
||
|
|
drawGraph(network.graph(1000, 800), '.svg');
|
||
|
|
```
|
||
|
|
|
||
|
|
will produce:
|
||
|
|
|
||
|
|
<img src="https://i.gyazo.com/a6a8076ce043f4892d0a77c6f816f0c0.png" width="100%"/>
|