Skip to content

Latest commit

 

History

History
89 lines (67 loc) · 3.2 KB

graph.md

File metadata and controls

89 lines (67 loc) · 3.2 KB
Raw

Class: Graph

The Graph class is important for the neural networks in a way that it keeps track of the relations between the matrices used in these networks. The following sections further describe the Graph class and its usage.

Class Structure

  • Constructor: Graph()
  • Provided Matrix Operations:
    • Delegate the actual Matrix Operation call
    • Keeps protocol of the sequence of matrix operations
    • Each operation returns a new Mat-Object, containing the specific results.
    • Available Matrix Operations are:
      • rowPluck(m: Mat, rowIndex: number): Mat
      • gauss(m: Mat, std: Mat): Mat
      • tanh(m: Mat): Mat
      • sig(m: Mat): Mat
      • relu(m: Mat): Mat
      • add(mat1: Mat, mat2: Mat): Mat
      • mul(mat1: Mat, mat2: Mat): Mat
      • dot(mat1: Mat, mat2: Mat): Mat
      • eltmul(mat1: Mat, mat2: Mat): Mat
  • memorizeOperationSequence(isMemorizing: boolean): void: Switch, whether the graph should keep a protocol of the operation sequence for backpropagation
  • isMemorizingSequence(): boolean: Get current memorization state
  • forgetCurrentSequence(): void: Clear the graph memory
  • backward(): void: Calls the Backpropagation Stack in reverse (LIFO) order of Matrix Operation Derivatives.

Usage

Object creation

Create a graph that does (or does not) memorize the sequence of Matrix Operations.

const graph = new Graph();

/* OPTIONAL: Set Backprop-state to `true` */
graph.memorizeOperationSequence(true);

Matrix Operation Call e.g. sig(m: Mat): Mat

Create a graph and inject a Mat object to call a sigmoid operation on its respective elements. A graph object - with backpropagation enabled - memorizes the derivatives of the matrix operations in the order they have been called.

const graph = new Graph();
/* OPTIONAL: Set Backprop-state to `true` */
graph.memorizeOperationSequence(true);

const mat = new Mat(4, 1);
mat.setFrom([0.1, 0.3, 0.9, 0.4]); /* fill Matrix with values */

/* 
 * - calls sigmoid operation on matrix,
 * - registers the matrix operation to the backpropagation stack (if activated) and
 * - returns a new matrix object with respective results
 */
const result = graph.sig(mat);

Call of the backward(): void Method

After the execution of a sequence of matrix operations via a Graph-object, this graph is then able to execute the backpropagation process in reverse (LIFO) order.

Prerequisite: The Graph-object needs to memorize the sequence of matrix operations.

  • Call graph.memorizeOperationSequence(true); for that
  • Check the memorization state of graph with graph.isMemorizingSequence();
const graph = new Graph();

const mat = new Mat(4, 1);
mat.setFrom([0.1, 0.3, 0.9, 0.4]); /* fill Matrix with values */

/* IMPORTANT: Set Backprop-state to `true` to be able to memorize matrix operation sequence */
graph.memorizeOperationSequence(true);

/* now: perform some sequence of matrix operations */
const result = graph.sig(mat);

/* manipulate the derivatives of `result`-Matrix */
result.dw[1] = -0.8;

/* propagate the modification back, by calling the memorized sequence of matrox operations */
graph.backward();

/* OPTIONAL: reset the graphs memory/protocol */
graph.forgetCurrentSequence();