add deep networks

2 files changed, 39 insertions, 12 deletions

diff --git a/src/mind/topology.js b/src/mind/topology.js
index 19eb399..320b499 100644
--- a/src/mind/topology.js
+++ b/src/mind/topology.js
@@ -91,7 +91,10 @@ function edge_ends(n, edge) {
 }
 
 
-function get_value(n, index, input) {
+function get_value(n, index, input, cache) {
+	if (cache !== undefined && cache[index]) {
+		return cache[index];
+	}
 	if (is_input(n, index)) {
 		return input[index];
 	}
@@ -102,13 +105,26 @@ function get_value(n, index, input) {
 		.map(x => edge_ends(n, x).source);
 
 	const sum = sources
-		.reduce((acc, x, i) => acc + (weight[i] * get_value(n, x, input)), 0);
+		.reduce((acc, x, i) => 
+			acc + (weight[i] * get_value(n, x, input, cache)), 
+			0
+		);
+
+	const value = Math.tanh(sum);
+
+	// !!! impure caching !!!
+	if (cache !== undefined) {
+		cache[index] = value;
+	}
 	
-	return Math.tanh(sum);
+	return value;
 }
 
 
 function network_compute(n, input, state) {
+	// !!! impure caching !!!
+	const value_cache = {};
+
 	const hidden = n.adjacency
 		.map((x, i) => 
 			(
@@ -121,11 +137,11 @@ function network_compute(n, input, state) {
 		.filter(i => i !== null);
 
 	const output = Object.freeze(
-		outputs.map(x => get_value(n, x, input))
+		outputs.map(x => get_value(n, x, input, value_cache))
 	);
 
 	const newstate = Object.freeze(
-		hidden.map(x => get_value(n, x, input))
+		hidden.map(x => get_value(n, x, input, value_cache))
 	);
 	
 	return Object.freeze([output, newstate]);
diff --git a/src/mind/topology.test.js b/src/mind/topology.test.js
index 7612c3d..5867763 100644
--- a/src/mind/topology.test.js
+++ b/src/mind/topology.test.js
@@ -169,10 +169,21 @@ test('hidden neurons', () => {
 });
 
 
-//test('arbitrary hidden neurons', () => {
-//	const n = network(1, 2, 1)
-//		.connect(0, 1, 1)
-//		.connect(1, 2, -1)
-//		.connect(2, 3, 2)
-//		.connect(3, 4, -2);
-//});
+test('arbitrary hidden neurons', () => {
+	const n = network(1, 2, 1)
+		.connect(0, 1, 1)
+		.connect(1, 2, -1)
+		.connect(2, 3, 2)
+
+	expect(n.compute([1], [0, 0])).toEqual([
+		[ Math.tanh (
+			2*Math.tanh( 
+				-1*Math.tanh( 1 ) 
+			)
+		) ],
+		[
+			Math.tanh( -Math.tanh(1) ),
+			Math.tanh(1),
+		],
+	]);
+});