@spiky-panda/core 0.1.0

package/dist/neuralnetwork/ann/mlp/mlp.builder.d.ts

@@ -0,0 +1,43 @@
+import { GraphNodeBuilder } from "../../../graph";
+import { ILayer } from "../../nn.interfaces";
+import { ILayerConnectionBuilder } from "../../nn.interfaces.builder";
+import { IActivationFunction, IMlpGraph, IMlpNeuron } from "./mlp.interfaces";
+export declare class MlpNeuronBuilder extends GraphNodeBuilder {
+    _bias?: number;
+    _activationFn?: IActivationFunction;
+    withBias(bias: number): MlpNeuronBuilder;
+    withActivationFn(fn: IActivationFunction): MlpNeuronBuilder;
+    build(...args: any[]): IMlpNeuron;
+}
+export declare class MlpLayerBuilder {
+    _count: number;
+    _neuronBuilder?: MlpNeuronBuilder;
+    withNeuron(bias: number | MlpNeuronBuilder, fn?: IActivationFunction): MlpLayerBuilder;
+    withCount(count: number): MlpLayerBuilder;
+    build(...args: any[]): ILayer<IMlpNeuron>;
+}
+export declare class PerceptronBuilder {
+    _inputLayerBuilder: MlpLayerBuilder;
+    _hiddenLayerBuilders: Array<MlpLayerBuilder>;
+    _outputLayerBuilder: MlpLayerBuilder;
+    _hiddenLayerCount: number;
+    _layerCount: number;
+    _defaultLayerConnBuilder: ILayerConnectionBuilder;
+    _connectionBuilders: Array<{
+        conn: ILayerConnectionBuilder;
+        from: number;
+        to: number;
+    }>;
+    withInputLayer(count: number | MlpLayerBuilder, bias?: number, activation?: IActivationFunction): PerceptronBuilder;
+    withHiddenLayer(count: number | MlpLayerBuilder | [{
+        count: number;
+        bias?: number;
+        activation?: IActivationFunction;
+        conn?: ILayerConnectionBuilder;
+    }], bias?: number, activation?: IActivationFunction): PerceptronBuilder;
+    withOutputLayer(count: number | MlpLayerBuilder, bias?: number, activation?: IActivationFunction): PerceptronBuilder;
+    withConnectionBuilder(connBuilder?: ILayerConnectionBuilder, from?: number, to?: number): PerceptronBuilder;
+    withDefaultConnectionBuilder(conn: ILayerConnectionBuilder): PerceptronBuilder;
+    build(...args: any[]): IMlpGraph;
+    reset(): PerceptronBuilder;
+}

package/dist/neuralnetwork/ann/mlp/mlp.builder.js

@@ -0,0 +1,168 @@
+import { GraphBuilder, GraphNodeBuilder } from "../../../graph";
+import { LayerConnectionBuilder } from "../../nn.builders";
+export class MlpNeuronBuilder extends GraphNodeBuilder {
+    withBias(bias) {
+        this._bias = bias;
+        return this;
+    }
+    withActivationFn(fn) {
+        this._activationFn = fn;
+        return this;
+    }
+    build(...args) {
+        if (this._bias == undefined) {
+            throw new Error("Bias must be provided.");
+        }
+        if (this._activationFn == undefined) {
+            throw new Error("Activation function must be provided.");
+        }
+        const neuron = super.build(...args);
+        neuron.bias = this._bias;
+        neuron.activationFn = this._activationFn;
+        return neuron;
+    }
+}
+export class MlpLayerBuilder {
+    constructor() {
+        this._count = 0;
+    }
+    withNeuron(bias, fn) {
+        if (bias instanceof MlpNeuronBuilder) {
+            this._neuronBuilder = bias;
+            return this;
+        }
+        this._neuronBuilder = this._neuronBuilder ?? new MlpNeuronBuilder();
+        this._neuronBuilder.withBias(bias);
+        if (fn) {
+            this._neuronBuilder.withActivationFn(fn);
+        }
+        return this;
+    }
+    withCount(count) {
+        this._count = count;
+        return this;
+    }
+    build(...args) {
+        if (!this._neuronBuilder) {
+            throw new Error("Neuron builder is not defined. Please set it using withNeuron() method.");
+        }
+        if (this._count <= 0) {
+            throw new Error("Neuron count must be greater than zero.");
+        }
+        const neurons = [];
+        for (let i = 0; i < this._count; i++) {
+            neurons.push(this._neuronBuilder.build(...args));
+        }
+        return neurons;
+    }
+}
+export class PerceptronBuilder {
+    constructor() {
+        this._inputLayerBuilder = new MlpLayerBuilder();
+        this._hiddenLayerBuilders = [];
+        this._outputLayerBuilder = new MlpLayerBuilder();
+        this._hiddenLayerCount = 0;
+        this._layerCount = 0;
+        this._defaultLayerConnBuilder = new LayerConnectionBuilder();
+        this._connectionBuilders = [];
+    }
+    withInputLayer(count, bias, activation) {
+        if (count instanceof MlpLayerBuilder) {
+            this._inputLayerBuilder = count;
+            this._layerCount++;
+            return this;
+        }
+        if (bias == undefined || activation == undefined) {
+            throw new Error("Bias and activation function must be provided when using count.");
+        }
+        this._inputLayerBuilder = this._inputLayerBuilder ?? new MlpLayerBuilder();
+        this._inputLayerBuilder.withCount(count).withNeuron(bias, activation);
+        this._layerCount++;
+        return this;
+    }
+    withHiddenLayer(count, bias, activation) {
+        if (count instanceof MlpLayerBuilder) {
+            this._hiddenLayerBuilders[this._hiddenLayerCount] = count;
+            this._layerCount++;
+            return this;
+        }
+        if (Array.isArray(count)) {
+            for (const layer of count) {
+                this.withHiddenLayer(layer.count, layer.bias, layer.activation);
+                if (layer.conn) {
+                    this.withConnectionBuilder(layer.conn);
+                }
+            }
+            return this;
+        }
+        if (bias == undefined || activation == undefined) {
+            throw new Error("Bias and activation function must be provided when using count.");
+        }
+        this._hiddenLayerBuilders[this._hiddenLayerCount] = this._hiddenLayerBuilders[this._hiddenLayerCount] ?? new MlpLayerBuilder();
+        this._hiddenLayerBuilders[this._hiddenLayerCount].withCount(count).withNeuron(bias, activation);
+        this._layerCount++;
+        return this;
+    }
+    withOutputLayer(count, bias, activation) {
+        if (count instanceof MlpLayerBuilder) {
+            this._outputLayerBuilder = count;
+            this._layerCount++;
+            return this;
+        }
+        if (bias == undefined || activation == undefined) {
+            throw new Error("Bias and activation function must be provided when using count.");
+        }
+        this._outputLayerBuilder = this._outputLayerBuilder ?? new MlpLayerBuilder();
+        this._outputLayerBuilder.withCount(count).withNeuron(bias, activation);
+        this._layerCount++;
+        return this;
+    }
+    withConnectionBuilder(connBuilder, from, to) {
+        const b = connBuilder ?? this._defaultLayerConnBuilder ?? new LayerConnectionBuilder();
+        const f = from != undefined ? from : this._layerCount - 2;
+        const t = to != undefined ? to : this._layerCount - 1;
+        if (f >= 0 && t >= 0) {
+            this._connectionBuilders.push({ conn: b, from: f, to: t });
+        }
+        return this;
+    }
+    withDefaultConnectionBuilder(conn) {
+        this._defaultLayerConnBuilder = conn;
+        return this;
+    }
+    build(...args) {
+        const builder = new GraphBuilder();
+        const layers = [];
+        layers.push(this._inputLayerBuilder.build(...args));
+        layers.push(...this._hiddenLayerBuilders.map((b) => b.build(...args)));
+        layers.push(this._outputLayerBuilder.build(...args));
+        for (let i = 0; i != this._connectionBuilders.length; i++) {
+            let layerIndex = this._connectionBuilders[i].from;
+            if (layerIndex < 0 || layerIndex >= layers.length) {
+                continue;
+            }
+            const a = layers[layerIndex];
+            layerIndex = this._connectionBuilders[i].to;
+            if (layerIndex < 0 || layerIndex >= layers.length) {
+                continue;
+            }
+            const b = layers[layerIndex];
+            const links = this._connectionBuilders[i].conn.build(a, b);
+            if (links)
+                builder.withLinks(links);
+        }
+        builder.withNodes(layers.flat());
+        const g = builder.build();
+        return g;
+    }
+    reset() {
+        this._inputLayerBuilder = new MlpLayerBuilder();
+        this._hiddenLayerBuilders = [];
+        this._outputLayerBuilder = new MlpLayerBuilder();
+        this._hiddenLayerCount = 0;
+        this._defaultLayerConnBuilder = new LayerConnectionBuilder();
+        this._connectionBuilders = [];
+        return this;
+    }
+}
+//# sourceMappingURL=mlp.builder.js.map

package/dist/neuralnetwork/ann/mlp/mlp.builder.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"mlp.builder.js","sourceRoot":"","sources":["../../../../src/neuralnetwork/ann/mlp/mlp.builder.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,gBAAgB,EAAE,MAAM,gBAAgB,CAAC;AAChE,OAAO,EAAE,sBAAsB,EAAE,MAAM,mBAAmB,CAAC;AAM3D,MAAM,OAAO,gBAAiB,SAAQ,gBAAgB;IAI3C,QAAQ,CAAC,IAAY;QACxB,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC;QAClB,OAAO,IAAI,CAAC;IAChB,CAAC;IACM,gBAAgB,CAAC,EAAuB;QAC3C,IAAI,CAAC,aAAa,GAAG,EAAE,CAAC;QACxB,OAAO,IAAI,CAAC;IAChB,CAAC;IACM,KAAK,CAAC,GAAG,IAAW;QACvB,IAAI,IAAI,CAAC,KAAK,IAAI,SAAS,EAAE,CAAC;YAC1B,MAAM,IAAI,KAAK,CAAC,wBAAwB,CAAC,CAAC;QAC9C,CAAC;QACD,IAAI,IAAI,CAAC,aAAa,IAAI,SAAS,EAAE,CAAC;YAClC,MAAM,IAAI,KAAK,CAAC,uCAAuC,CAAC,CAAC;QAC7D,CAAC;QACD,MAAM,MAAM,GAAG,KAAK,CAAC,KAAK,CAAC,GAAG,IAAI,CAAe,CAAC;QAClD,MAAM,CAAC,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC;QACzB,MAAM,CAAC,YAAY,GAAG,IAAI,CAAC,aAAa,CAAC;QACzC,OAAO,MAAM,CAAC;IAClB,CAAC;CACJ;AAED,MAAM,OAAO,eAAe;IAA5B;QACI,WAAM,GAAW,CAAC,CAAC;IAkCvB,CAAC;IA/BU,UAAU,CAAC,IAA+B,EAAE,EAAwB;QACvE,IAAI,IAAI,YAAY,gBAAgB,EAAE,CAAC;YACnC,IAAI,CAAC,cAAc,GAAG,IAAI,CAAC;YAC3B,OAAO,IAAI,CAAC;QAChB,CAAC;QACD,IAAI,CAAC,cAAc,GAAG,IAAI,CAAC,cAAc,IAAI,IAAI,gBAAgB,EAAE,CAAC;QACpE,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;QACnC,IAAI,EAAE,EAAE,CAAC;YACL,IAAI,CAAC,cAAc,CAAC,gBAAgB,CAAC,EAAE,CAAC,CAAC;QAC7C,CAAC;QACD,OAAO,IAAI,CAAC;IAChB,CAAC;IAEM,SAAS,CAAC,KAAa;QAC1B,IAAI,CAAC,MAAM,GAAG,KAAK,CAAC;QACpB,OAAO,IAAI,CAAC;IAChB,CAAC;IAEM,KAAK,CAAC,GAAG,IAAW;QACvB,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,CAAC;YACvB,MAAM,IAAI,KAAK,CAAC,yEAAyE,CAAC,CAAC;QAC/F,CAAC;QACD,IAAI,IAAI,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;YACnB,MAAM,IAAI,KAAK,CAAC,yCAAyC,CAAC,CAAC;QAC/D,CAAC;QACD,MAAM,OAAO,GAAuB,EAAE,CAAC;QACvC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YACnC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,cAAc,CAAC,KAAK,CAAC,GAAG,IAAI,CAAC,CAAC,CAAC;QACrD,CAAC;QACD,OAAO,OAAO,CAAC;IACnB,CAAC;CACJ;AAED,MAAM,OAAO,iBAAiB;IAA9B;QACI,uBAAkB,GAAoB,IAAI,eAAe,EAAE,CAAC;QAC5D,yBAAoB,GAA2B,EAAE,CAAC;QAClD,wBAAmB,GAAoB,IAAI,eAAe,EAAE,CAAC;QAC7D,sBAAiB,GAAW,CAAC,CAAC;QAC9B,gBAAW,GAAW,CAAC,CAAC;QACxB,6BAAwB,GAA4B,IAAI,sBAAsB,EAAE,CAAC;QACjF,wBAAmB,GAAuE,EAAE,CAAC;IAiHjG,CAAC;IA/GU,cAAc,CAAC,KAA+B,EAAE,IAAa,EAAE,UAAgC;QAClG,IAAI,KAAK,YAAY,eAAe,EAAE,CAAC;YACnC,IAAI,CAAC,kBAAkB,GAAG,KAAK,CAAC;YAChC,IAAI,CAAC,WAAW,EAAE,CAAC;YACnB,OAAO,IAAI,CAAC;QAChB,CAAC;QACD,IAAI,IAAI,IAAI,SAAS,IAAI,UAAU,IAAI,SAAS,EAAE,CAAC;YAC/C,MAAM,IAAI,KAAK,CAAC,iEAAiE,CAAC,CAAC;QACvF,CAAC;QACD,IAAI,CAAC,kBAAkB,GAAG,IAAI,CAAC,kBAAkB,IAAI,IAAI,eAAe,EAAE,CAAC;QAC3E,IAAI,CAAC,kBAAkB,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,UAAU,CAAC,IAAI,EAAE,UAAU,CAAC,CAAC;QACtE,IAAI,CAAC,WAAW,EAAE,CAAC;QACnB,OAAO,IAAI,CAAC;IAChB,CAAC;IAEM,eAAe,CAClB,KAAsI,EACtI,IAAa,EACb,UAAgC;QAEhC,IAAI,KAAK,YAAY,eAAe,EAAE,CAAC;YACnC,IAAI,CAAC,oBAAoB,CAAC,IAAI,CAAC,iBAAiB,CAAC,GAAG,KAAK,CAAC;YAC1D,IAAI,CAAC,WAAW,EAAE,CAAC;YACnB,OAAO,IAAI,CAAC;QAChB,CAAC;QACD,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC;YACvB,KAAK,MAAM,KAAK,IAAI,KAAK,EAAE,CAAC;gBACxB,IAAI,CAAC,eAAe,CAAC,KAAK,CAAC,KAAK,EAAE,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,UAAU,CAAC,CAAC;gBAChE,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC;oBACb,IAAI,CAAC,qBAAqB,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;gBAC3C,CAAC;YACL,CAAC;YACD,OAAO,IAAI,CAAC;QAChB,CAAC;QACD,IAAI,IAAI,IAAI,SAAS,IAAI,UAAU,IAAI,SAAS,EAAE,CAAC;YAC/C,MAAM,IAAI,KAAK,CAAC,iEAAiE,CAAC,CAAC;QACvF,CAAC;QACD,IAAI,CAAC,oBAAoB,CAAC,IAAI,CAAC,iBAAiB,CAAC,GAAG,IAAI,CAAC,oBAAoB,CAAC,IAAI,CAAC,iBAAiB,CAAC,IAAI,IAAI,eAAe,EAAE,CAAC;QAC/H,IAAI,CAAC,oBAAoB,CAAC,IAAI,CAAC,iBAAiB,CAAC,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,UAAU,CAAC,IAAI,EAAE,UAAU,CAAC,CAAC;QAChG,IAAI,CAAC,WAAW,EAAE,CAAC;QACnB,OAAO,IAAI,CAAC;IAChB,CAAC;IAEM,eAAe,CAAC,KAA+B,EAAE,IAAa,EAAE,UAAgC;QACnG,IAAI,KAAK,YAAY,eAAe,EAAE,CAAC;YACnC,IAAI,CAAC,mBAAmB,GAAG,KAAK,CAAC;YACjC,IAAI,CAAC,WAAW,EAAE,CAAC;YACnB,OAAO,IAAI,CAAC;QAChB,CAAC;QACD,IAAI,IAAI,IAAI,SAAS,IAAI,UAAU,IAAI,SAAS,EAAE,CAAC;YAC/C,MAAM,IAAI,KAAK,CAAC,iEAAiE,CAAC,CAAC;QACvF,CAAC;QACD,IAAI,CAAC,mBAAmB,GAAG,IAAI,CAAC,mBAAmB,IAAI,IAAI,eAAe,EAAE,CAAC;QAC7E,IAAI,CAAC,mBAAmB,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,UAAU,CAAC,IAAI,EAAE,UAAU,CAAC,CAAC;QACvE,IAAI,CAAC,WAAW,EAAE,CAAC;QACnB,OAAO,IAAI,CAAC;IAChB,CAAC;IAEM,qBAAqB,CAAC,WAAqC,EAAE,IAAa,EAAE,EAAW;QAC1F,MAAM,CAAC,GAAG,WAAW,IAAI,IAAI,CAAC,wBAAwB,IAAI,IAAI,sBAAsB,EAAE,CAAC;QACvF,MAAM,CAAC,GAAG,IAAI,IAAI,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,WAAW,GAAG,CAAC,CAAC;QAC1D,MAAM,CAAC,GAAG,EAAE,IAAI,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,WAAW,GAAG,CAAC,CAAC;QACtD,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;YACnB,IAAI,CAAC,mBAAmB,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,CAAC,EAAE,IAAI,EAAE,CAAC,EAAE,EAAE,EAAE,CAAC,EAAE,CAAC,CAAC;QAC/D,CAAC;QACD,OAAO,IAAI,CAAC;IAChB,CAAC;IAEM,4BAA4B,CAAC,IAA6B;QAC7D,IAAI,CAAC,wBAAwB,GAAG,IAAI,CAAC;QACrC,OAAO,IAAI,CAAC;IAChB,CAAC;IAEM,KAAK,CAAC,GAAG,IAAW;QACvB,MAAM,OAAO,GAAG,IAAI,YAAY,EAAE,CAAC;QAEnC,MAAM,MAAM,GAAG,EAAE,CAAC;QAClB,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,kBAAkB,CAAC,KAAK,CAAC,GAAG,IAAI,CAAC,CAAC,CAAC;QACpD,MAAM,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC,oBAAoB,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC;QACvE,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,mBAAmB,CAAC,KAAK,CAAC,GAAG,IAAI,CAAC,CAAC,CAAC;QAErD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,IAAI,CAAC,mBAAmB,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YACxD,IAAI,UAAU,GAAG,IAAI,CAAC,mBAAmB,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;YAClD,IAAI,UAAU,GAAG,CAAC,IAAI,UAAU,IAAI,MAAM,CAAC,MAAM,EAAE,CAAC;gBAChD,SAAS;YACb,CAAC;YACD,MAAM,CAAC,GAAG,MAAM,CAAC,UAAU,CAAC,CAAC;YAC7B,UAAU,GAAG,IAAI,CAAC,mBAAmB,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;YAC5C,IAAI,UAAU,GAAG,CAAC,IAAI,UAAU,IAAI,MAAM,CAAC,MAAM,EAAE,CAAC;gBAChD,SAAS;YACb,CAAC;YACD,MAAM,CAAC,GAAG,MAAM,CAAC,UAAU,CAAC,CAAC;YAC7B,MAAM,KAAK,GAAG,IAAI,CAAC,mBAAmB,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;YAC3D,IAAI,KAAK;gBAAE,OAAO,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;QACxC,CAAC;QAED,OAAO,CAAC,SAAS,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC;QAEjC,MAAM,CAAC,GAAG,OAAO,CAAC,KAAK,EAAe,CAAC;QACvC,OAAO,CAAC,CAAC;IACb,CAAC;IAEM,KAAK;QACR,IAAI,CAAC,kBAAkB,GAAG,IAAI,eAAe,EAAE,CAAC;QAChD,IAAI,CAAC,oBAAoB,GAAG,EAAE,CAAC;QAC/B,IAAI,CAAC,mBAAmB,GAAG,IAAI,eAAe,EAAE,CAAC;QACjD,IAAI,CAAC,iBAAiB,GAAG,CAAC,CAAC;QAC3B,IAAI,CAAC,wBAAwB,GAAG,IAAI,sBAAsB,EAAE,CAAC;QAC7D,IAAI,CAAC,mBAAmB,GAAG,EAAE,CAAC;QAC9B,OAAO,IAAI,CAAC;IAChB,CAAC;CACJ","sourcesContent":["import { GraphBuilder, GraphNodeBuilder } from \"../../../graph\";\r\nimport { LayerConnectionBuilder } from \"../../nn.builders\";\r\nimport { ILayer } from \"../../nn.interfaces\";\r\nimport { ILayerConnectionBuilder } from \"../../nn.interfaces.builder\";\r\n\r\nimport { IActivationFunction, IMlpGraph, IMlpNeuron } from \"./mlp.interfaces\";\r\n\r\nexport class MlpNeuronBuilder extends GraphNodeBuilder {\r\n    _bias?: number;\r\n    _activationFn?: IActivationFunction;\r\n\r\n    public withBias(bias: number): MlpNeuronBuilder {\r\n        this._bias = bias;\r\n        return this;\r\n    }\r\n    public withActivationFn(fn: IActivationFunction): MlpNeuronBuilder {\r\n        this._activationFn = fn;\r\n        return this;\r\n    }\r\n    public build(...args: any[]): IMlpNeuron {\r\n        if (this._bias == undefined) {\r\n            throw new Error(\"Bias must be provided.\");\r\n        }\r\n        if (this._activationFn == undefined) {\r\n            throw new Error(\"Activation function must be provided.\");\r\n        }\r\n        const neuron = super.build(...args) as IMlpNeuron;\r\n        neuron.bias = this._bias;\r\n        neuron.activationFn = this._activationFn;\r\n        return neuron;\r\n    }\r\n}\r\n\r\nexport class MlpLayerBuilder {\r\n    _count: number = 0;\r\n    _neuronBuilder?: MlpNeuronBuilder;\r\n\r\n    public withNeuron(bias: number | MlpNeuronBuilder, fn?: IActivationFunction): MlpLayerBuilder {\r\n        if (bias instanceof MlpNeuronBuilder) {\r\n            this._neuronBuilder = bias;\r\n            return this;\r\n        }\r\n        this._neuronBuilder = this._neuronBuilder ?? new MlpNeuronBuilder();\r\n        this._neuronBuilder.withBias(bias);\r\n        if (fn) {\r\n            this._neuronBuilder.withActivationFn(fn);\r\n        }\r\n        return this;\r\n    }\r\n\r\n    public withCount(count: number): MlpLayerBuilder {\r\n        this._count = count;\r\n        return this;\r\n    }\r\n\r\n    public build(...args: any[]): ILayer<IMlpNeuron> {\r\n        if (!this._neuronBuilder) {\r\n            throw new Error(\"Neuron builder is not defined. Please set it using withNeuron() method.\");\r\n        }\r\n        if (this._count <= 0) {\r\n            throw new Error(\"Neuron count must be greater than zero.\");\r\n        }\r\n        const neurons: ILayer<IMlpNeuron> = [];\r\n        for (let i = 0; i < this._count; i++) {\r\n            neurons.push(this._neuronBuilder.build(...args));\r\n        }\r\n        return neurons;\r\n    }\r\n}\r\n\r\nexport class PerceptronBuilder {\r\n    _inputLayerBuilder: MlpLayerBuilder = new MlpLayerBuilder();\r\n    _hiddenLayerBuilders: Array<MlpLayerBuilder> = [];\r\n    _outputLayerBuilder: MlpLayerBuilder = new MlpLayerBuilder();\r\n    _hiddenLayerCount: number = 0;\r\n    _layerCount: number = 0;\r\n    _defaultLayerConnBuilder: ILayerConnectionBuilder = new LayerConnectionBuilder();\r\n    _connectionBuilders: Array<{ conn: ILayerConnectionBuilder; from: number; to: number }> = [];\r\n\r\n    public withInputLayer(count: number | MlpLayerBuilder, bias?: number, activation?: IActivationFunction): PerceptronBuilder {\r\n        if (count instanceof MlpLayerBuilder) {\r\n            this._inputLayerBuilder = count;\r\n            this._layerCount++;\r\n            return this;\r\n        }\r\n        if (bias == undefined || activation == undefined) {\r\n            throw new Error(\"Bias and activation function must be provided when using count.\");\r\n        }\r\n        this._inputLayerBuilder = this._inputLayerBuilder ?? new MlpLayerBuilder();\r\n        this._inputLayerBuilder.withCount(count).withNeuron(bias, activation);\r\n        this._layerCount++;\r\n        return this;\r\n    }\r\n\r\n    public withHiddenLayer(\r\n        count: number | MlpLayerBuilder | [{ count: number; bias?: number; activation?: IActivationFunction; conn?: ILayerConnectionBuilder }],\r\n        bias?: number,\r\n        activation?: IActivationFunction\r\n    ): PerceptronBuilder {\r\n        if (count instanceof MlpLayerBuilder) {\r\n            this._hiddenLayerBuilders[this._hiddenLayerCount] = count;\r\n            this._layerCount++;\r\n            return this;\r\n        }\r\n        if (Array.isArray(count)) {\r\n            for (const layer of count) {\r\n                this.withHiddenLayer(layer.count, layer.bias, layer.activation);\r\n                if (layer.conn) {\r\n                    this.withConnectionBuilder(layer.conn);\r\n                }\r\n            }\r\n            return this;\r\n        }\r\n        if (bias == undefined || activation == undefined) {\r\n            throw new Error(\"Bias and activation function must be provided when using count.\");\r\n        }\r\n        this._hiddenLayerBuilders[this._hiddenLayerCount] = this._hiddenLayerBuilders[this._hiddenLayerCount] ?? new MlpLayerBuilder();\r\n        this._hiddenLayerBuilders[this._hiddenLayerCount].withCount(count).withNeuron(bias, activation);\r\n        this._layerCount++;\r\n        return this;\r\n    }\r\n\r\n    public withOutputLayer(count: number | MlpLayerBuilder, bias?: number, activation?: IActivationFunction): PerceptronBuilder {\r\n        if (count instanceof MlpLayerBuilder) {\r\n            this._outputLayerBuilder = count;\r\n            this._layerCount++;\r\n            return this;\r\n        }\r\n        if (bias == undefined || activation == undefined) {\r\n            throw new Error(\"Bias and activation function must be provided when using count.\");\r\n        }\r\n        this._outputLayerBuilder = this._outputLayerBuilder ?? new MlpLayerBuilder();\r\n        this._outputLayerBuilder.withCount(count).withNeuron(bias, activation);\r\n        this._layerCount++;\r\n        return this;\r\n    }\r\n\r\n    public withConnectionBuilder(connBuilder?: ILayerConnectionBuilder, from?: number, to?: number): PerceptronBuilder {\r\n        const b = connBuilder ?? this._defaultLayerConnBuilder ?? new LayerConnectionBuilder();\r\n        const f = from != undefined ? from : this._layerCount - 2;\r\n        const t = to != undefined ? to : this._layerCount - 1;\r\n        if (f >= 0 && t >= 0) {\r\n            this._connectionBuilders.push({ conn: b, from: f, to: t });\r\n        }\r\n        return this;\r\n    }\r\n\r\n    public withDefaultConnectionBuilder(conn: ILayerConnectionBuilder): PerceptronBuilder {\r\n        this._defaultLayerConnBuilder = conn;\r\n        return this;\r\n    }\r\n\r\n    public build(...args: any[]): IMlpGraph {\r\n        const builder = new GraphBuilder();\r\n\r\n        const layers = [];\r\n        layers.push(this._inputLayerBuilder.build(...args));\r\n        layers.push(...this._hiddenLayerBuilders.map((b) => b.build(...args)));\r\n        layers.push(this._outputLayerBuilder.build(...args));\r\n\r\n        for (let i = 0; i != this._connectionBuilders.length; i++) {\r\n            let layerIndex = this._connectionBuilders[i].from;\r\n            if (layerIndex < 0 || layerIndex >= layers.length) {\r\n                continue;\r\n            }\r\n            const a = layers[layerIndex];\r\n            layerIndex = this._connectionBuilders[i].to;\r\n            if (layerIndex < 0 || layerIndex >= layers.length) {\r\n                continue;\r\n            }\r\n            const b = layers[layerIndex];\r\n            const links = this._connectionBuilders[i].conn.build(a, b);\r\n            if (links) builder.withLinks(links);\r\n        }\r\n\r\n        builder.withNodes(layers.flat());\r\n\r\n        const g = builder.build() as IMlpGraph;\r\n        return g;\r\n    }\r\n\r\n    public reset(): PerceptronBuilder {\r\n        this._inputLayerBuilder = new MlpLayerBuilder();\r\n        this._hiddenLayerBuilders = [];\r\n        this._outputLayerBuilder = new MlpLayerBuilder();\r\n        this._hiddenLayerCount = 0;\r\n        this._defaultLayerConnBuilder = new LayerConnectionBuilder();\r\n        this._connectionBuilders = [];\r\n        return this;\r\n    }\r\n}\r\n"]}

package/dist/neuralnetwork/ann/mlp/mlp.graph.d.ts

@@ -0,0 +1,7 @@
+import { ICartesian } from "../../../geometry";
+import { Graph } from "../../../graph";
+import { Nullable } from "../../../types";
+import { IMlpGraph, IMlpNeuron, IMlpSynapse } from "./mlp.interfaces";
+export declare class MlpGraph extends Graph<IMlpNeuron, IMlpSynapse> implements IMlpGraph {
+    constructor(nodes?: IMlpNeuron[], links?: IMlpSynapse[], inputs?: Nullable<IMlpNeuron[]>, outputs?: Nullable<IMlpNeuron[]>, hiddens?: Nullable<IMlpNeuron[]>, onsc?: Nullable<IMlpSynapse[]>, opsc?: Nullable<IMlpSynapse[]>, position?: ICartesian);
+}

package/dist/neuralnetwork/ann/mlp/mlp.graph.js

@@ -0,0 +1,7 @@
+import { Graph } from "../../../graph";
+export class MlpGraph extends Graph {
+    constructor(nodes = [], links = [], inputs = null, outputs = null, hiddens = null, onsc = null, opsc = null, position) {
+        super(nodes, links, inputs, outputs, hiddens, onsc, opsc, position);
+    }
+}
+//# sourceMappingURL=mlp.graph.js.map

package/dist/neuralnetwork/ann/mlp/mlp.graph.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"mlp.graph.js","sourceRoot":"","sources":["../../../../src/neuralnetwork/ann/mlp/mlp.graph.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,KAAK,EAAE,MAAM,gBAAgB,CAAC;AAIvC,MAAM,OAAO,QAAS,SAAQ,KAA8B;IACxD,YACI,QAAsB,EAAE,EACxB,QAAuB,EAAE,EACzB,SAAiC,IAAI,EACrC,UAAkC,IAAI,EACtC,UAAkC,IAAI,EACtC,OAAgC,IAAI,EACpC,OAAgC,IAAI,EACpC,QAAqB;QAErB,KAAK,CAAC,KAAK,EAAE,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,IAAI,EAAE,QAAQ,CAAC,CAAC;IACxE,CAAC;CACJ","sourcesContent":["import { ICartesian } from \"../../../geometry\";\r\nimport { Graph } from \"../../../graph\";\r\nimport { Nullable } from \"../../../types\";\r\nimport { IMlpGraph, IMlpNeuron, IMlpSynapse } from \"./mlp.interfaces\";\r\n\r\nexport class MlpGraph extends Graph<IMlpNeuron, IMlpSynapse> implements IMlpGraph {\r\n    public constructor(\r\n        nodes: IMlpNeuron[] = [],\r\n        links: IMlpSynapse[] = [],\r\n        inputs: Nullable<IMlpNeuron[]> = null,\r\n        outputs: Nullable<IMlpNeuron[]> = null,\r\n        hiddens: Nullable<IMlpNeuron[]> = null,\r\n        onsc: Nullable<IMlpSynapse[]> = null,\r\n        opsc: Nullable<IMlpSynapse[]> = null,\r\n        position?: ICartesian\r\n    ) {\r\n        super(nodes, links, inputs, outputs, hiddens, onsc, opsc, position);\r\n    }\r\n}\r\n"]}

package/dist/neuralnetwork/ann/mlp/mlp.inference.d.ts

@@ -0,0 +1,13 @@
+import { IMlpGraph, IActivationFunction } from "./mlp.interfaces";
+export declare class MLPInferenceRuntime {
+    readonly graph: IMlpGraph;
+    mainActivation: IActivationFunction;
+    constructor(graph: IMlpGraph, mainActivation?: IActivationFunction);
+    /**
+     * Runs inference on the MLP graph given an array of input values.
+     * The number of input values must match the number of input neurons.
+     */
+    run(inputValues: number[]): number[];
+    clearContext(): void;
+    deleteContext(): void;
+}

package/dist/neuralnetwork/ann/mlp/mlp.inference.js

@@ -0,0 +1,63 @@
+import { ActivationFunctions } from "./mlp.activation";
+import { MLPRuntimeUtils } from "./mlp.runtime.utils";
+export class MLPInferenceRuntime {
+    constructor(graph, mainActivation = ActivationFunctions.relu) {
+        this.graph = graph;
+        this.mainActivation = mainActivation;
+    }
+    /**
+     * Runs inference on the MLP graph given an array of input values.
+     * The number of input values must match the number of input neurons.
+     */
+    run(inputValues) {
+        if (inputValues.length !== this.graph.inputs.length) {
+            throw new Error(`Input count mismatch: expected ${this.graph.inputs.length}, got ${inputValues.length}`);
+        }
+        const ready = [];
+        // Initialize context for all neurons
+        for (const neuron of this.graph.nodes) {
+            MLPRuntimeUtils.resetInferenceContext(neuron);
+        }
+        // Load input values and mark as ready
+        for (let i = 0; i < inputValues.length; i++) {
+            const neuron = this.graph.inputs[i];
+            const ctx = neuron.bag;
+            ctx.sum = inputValues[i];
+            ctx.activation = inputValues[i];
+            ctx.remainingInputs = 0;
+            ready.push(neuron);
+        }
+        // Process forward pass
+        while (ready.length > 0) {
+            const source = ready.shift();
+            const sourceCtx = source.bag;
+            const outputs = source.onsc() ?? [];
+            for (const syn of outputs) {
+                const target = syn.ofin;
+                const targetCtx = target.bag;
+                targetCtx.sum += sourceCtx.activation * syn.weight;
+                targetCtx.remainingInputs--;
+                if (targetCtx.remainingInputs === 0) {
+                    targetCtx.sum += target.bias;
+                    const afn = target.activationFn ?? this.mainActivation;
+                    targetCtx.activation = afn.fn(targetCtx.sum);
+                    ready.push(target);
+                }
+            }
+        }
+        // Return output activations
+        const results = this.graph.outputs.map((n) => n.bag.activation);
+        return results;
+    }
+    clearContext() {
+        for (const neuron of this.graph.nodes) {
+            MLPRuntimeUtils.resetInferenceContext(neuron);
+        }
+    }
+    deleteContext() {
+        for (const neuron of this.graph.nodes) {
+            neuron.bag = undefined;
+        }
+    }
+}
+//# sourceMappingURL=mlp.inference.js.map

package/dist/neuralnetwork/ann/mlp/mlp.inference.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"mlp.inference.js","sourceRoot":"","sources":["../../../../src/neuralnetwork/ann/mlp/mlp.inference.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,mBAAmB,EAAE,MAAM,kBAAkB,CAAC;AAEvD,OAAO,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAC;AAEtD,MAAM,OAAO,mBAAmB;IAC5B,YAAmC,KAAgB,EAAS,iBAAsC,mBAAmB,CAAC,IAAI;QAAvF,UAAK,GAAL,KAAK,CAAW;QAAS,mBAAc,GAAd,cAAc,CAAgD;IAAG,CAAC;IAE9H;;;OAGG;IACH,GAAG,CAAC,WAAqB;QACrB,IAAI,WAAW,CAAC,MAAM,KAAK,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC;YAClD,MAAM,IAAI,KAAK,CAAC,kCAAkC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,SAAS,WAAW,CAAC,MAAM,EAAE,CAAC,CAAC;QAC7G,CAAC;QAED,MAAM,KAAK,GAAiB,EAAE,CAAC;QAE/B,qCAAqC;QACrC,KAAK,MAAM,MAAM,IAAI,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,CAAC;YACpC,eAAe,CAAC,qBAAqB,CAAC,MAAM,CAAC,CAAC;QAClD,CAAC;QAED,sCAAsC;QACtC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,WAAW,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YAC1C,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;YACpC,MAAM,GAAG,GAAG,MAAM,CAAC,GAA8B,CAAC;YAElD,GAAG,CAAC,GAAG,GAAG,WAAW,CAAC,CAAC,CAAC,CAAC;YACzB,GAAG,CAAC,UAAU,GAAG,WAAW,CAAC,CAAC,CAAC,CAAC;YAChC,GAAG,CAAC,eAAe,GAAG,CAAC,CAAC;YACxB,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACvB,CAAC;QAED,uBAAuB;QACvB,OAAO,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACtB,MAAM,MAAM,GAAG,KAAK,CAAC,KAAK,EAAG,CAAC;YAC9B,MAAM,SAAS,GAAG,MAAM,CAAC,GAA8B,CAAC;YAExD,MAAM,OAAO,GAAG,MAAM,CAAC,IAAI,EAAe,IAAI,EAAE,CAAC;YACjD,KAAK,MAAM,GAAG,IAAI,OAAO,EAAE,CAAC;gBACxB,MAAM,MAAM,GAAG,GAAG,CAAC,IAAkB,CAAC;gBACtC,MAAM,SAAS,GAAG,MAAM,CAAC,GAA8B,CAAC;gBAExD,SAAS,CAAC,GAAG,IAAI,SAAS,CAAC,UAAU,GAAG,GAAG,CAAC,MAAM,CAAC;gBACnD,SAAS,CAAC,eAAe,EAAE,CAAC;gBAE5B,IAAI,SAAS,CAAC,eAAe,KAAK,CAAC,EAAE,CAAC;oBAClC,SAAS,CAAC,GAAG,IAAI,MAAM,CAAC,IAAI,CAAC;oBAC7B,MAAM,GAAG,GAAG,MAAM,CAAC,YAAY,IAAI,IAAI,CAAC,cAAc,CAAC;oBACvD,SAAS,CAAC,UAAU,GAAG,GAAG,CAAC,EAAE,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC;oBAC7C,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;gBACvB,CAAC;YACL,CAAC;QACL,CAAC;QAED,4BAA4B;QAC5B,MAAM,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAE,CAAC,CAAC,GAA+B,CAAC,UAAU,CAAC,CAAC;QAC7F,OAAO,OAAO,CAAC;IACnB,CAAC;IAEM,YAAY;QACf,KAAK,MAAM,MAAM,IAAI,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,CAAC;YACpC,eAAe,CAAC,qBAAqB,CAAC,MAAM,CAAC,CAAC;QAClD,CAAC;IACL,CAAC;IACM,aAAa;QAChB,KAAK,MAAM,MAAM,IAAI,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,CAAC;YACpC,MAAM,CAAC,GAAG,GAAG,SAAS,CAAC;QAC3B,CAAC;IACL,CAAC;CACJ","sourcesContent":["import { ActivationFunctions } from \"./mlp.activation\";\r\nimport { IMlpGraph, IMlpNeuron, IMlpSynapse, IInferenceNeuronContext, IActivationFunction } from \"./mlp.interfaces\";\r\nimport { MLPRuntimeUtils } from \"./mlp.runtime.utils\";\r\n\r\nexport class MLPInferenceRuntime {\r\n    public constructor(public readonly graph: IMlpGraph, public mainActivation: IActivationFunction = ActivationFunctions.relu) {}\r\n\r\n    /**\r\n     * Runs inference on the MLP graph given an array of input values.\r\n     * The number of input values must match the number of input neurons.\r\n     */\r\n    run(inputValues: number[]): number[] {\r\n        if (inputValues.length !== this.graph.inputs.length) {\r\n            throw new Error(`Input count mismatch: expected ${this.graph.inputs.length}, got ${inputValues.length}`);\r\n        }\r\n\r\n        const ready: IMlpNeuron[] = [];\r\n\r\n        // Initialize context for all neurons\r\n        for (const neuron of this.graph.nodes) {\r\n            MLPRuntimeUtils.resetInferenceContext(neuron);\r\n        }\r\n\r\n        // Load input values and mark as ready\r\n        for (let i = 0; i < inputValues.length; i++) {\r\n            const neuron = this.graph.inputs[i];\r\n            const ctx = neuron.bag as IInferenceNeuronContext;\r\n\r\n            ctx.sum = inputValues[i];\r\n            ctx.activation = inputValues[i];\r\n            ctx.remainingInputs = 0;\r\n            ready.push(neuron);\r\n        }\r\n\r\n        // Process forward pass\r\n        while (ready.length > 0) {\r\n            const source = ready.shift()!;\r\n            const sourceCtx = source.bag as IInferenceNeuronContext;\r\n\r\n            const outputs = source.onsc<IMlpSynapse>() ?? [];\r\n            for (const syn of outputs) {\r\n                const target = syn.ofin as IMlpNeuron;\r\n                const targetCtx = target.bag as IInferenceNeuronContext;\r\n\r\n                targetCtx.sum += sourceCtx.activation * syn.weight;\r\n                targetCtx.remainingInputs--;\r\n\r\n                if (targetCtx.remainingInputs === 0) {\r\n                    targetCtx.sum += target.bias;\r\n                    const afn = target.activationFn ?? this.mainActivation;\r\n                    targetCtx.activation = afn.fn(targetCtx.sum);\r\n                    ready.push(target);\r\n                }\r\n            }\r\n        }\r\n\r\n        // Return output activations\r\n        const results = this.graph.outputs.map((n) => (n.bag as IInferenceNeuronContext).activation);\r\n        return results;\r\n    }\r\n\r\n    public clearContext() {\r\n        for (const neuron of this.graph.nodes) {\r\n            MLPRuntimeUtils.resetInferenceContext(neuron);\r\n        }\r\n    }\r\n    public deleteContext() {\r\n        for (const neuron of this.graph.nodes) {\r\n            neuron.bag = undefined;\r\n        }\r\n    }\r\n}\r\n"]}

package/dist/neuralnetwork/ann/mlp/mlp.interfaces.d.ts

@@ -0,0 +1,21 @@
+import { IGraph } from "../../../graph";
+import { INeuron, ISynapse } from "../../nn.interfaces";
+export interface IActivationFunction {
+    fn: (x: number) => number;
+    derivative: (y: number) => number;
+}
+export interface IInferenceNeuronContext {
+    sum: number;
+    activation: number;
+    remainingInputs: number;
+    totalInputs: number;
+}
+export interface IMlpNeuron extends INeuron {
+    bias: number;
+    activationFn?: IActivationFunction;
+}
+export declare function isMlpNeuron(obj: unknown): obj is IMlpNeuron;
+export interface IMlpSynapse extends ISynapse {
+}
+export interface IMlpGraph extends IGraph<IMlpNeuron, IMlpSynapse> {
+}

package/dist/neuralnetwork/ann/mlp/mlp.interfaces.js

@@ -0,0 +1,7 @@
+/// <summary>
+/// Type guard to check whether an object is an IMlpNeuron
+/// </summary>
+export function isMlpNeuron(obj) {
+    return typeof obj === "object" && obj !== null && "bias" in obj && "activationFn" in obj;
+}
+//# sourceMappingURL=mlp.interfaces.js.map

package/dist/neuralnetwork/ann/mlp/mlp.interfaces.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"mlp.interfaces.js","sourceRoot":"","sources":["../../../../src/neuralnetwork/ann/mlp/mlp.interfaces.ts"],"names":[],"mappings":"AA8BA,aAAa;AACb,0DAA0D;AAC1D,cAAc;AACd,MAAM,UAAU,WAAW,CAAC,GAAY;IACpC,OAAO,OAAO,GAAG,KAAK,QAAQ,IAAI,GAAG,KAAK,IAAI,IAAI,MAAM,IAAI,GAAG,IAAI,cAAc,IAAI,GAAG,CAAC;AAC7F,CAAC","sourcesContent":["import { IGraph } from \"../../../graph\";\r\nimport { INeuron, ISynapse } from \"../../nn.interfaces\";\r\n\r\n/// <summary>\r\n/// Represents an activation function and its derivative.\r\n/// The derivative is expected to receive the activation output (f(x)), not the pre-activation input (x).\r\n/// </summary>\r\nexport interface IActivationFunction {\r\n    fn: (x: number) => number;\r\n    derivative: (y: number) => number;\r\n}\r\n\r\nexport interface IInferenceNeuronContext {\r\n    sum: number;\r\n    activation: number;\r\n    remainingInputs: number;\r\n    totalInputs: number;\r\n}\r\n\r\n/// <summary>\r\n/// Represents a neuron in a Multi-Layer Perceptron (MLP).\r\n/// It exposes its own state directly and includes a configurable bias.\r\n/// </summary>\r\nexport interface IMlpNeuron extends INeuron {\r\n    /// <summary>Bias added before applying the activation function</summary>\r\n    bias: number;\r\n    /// <summary>Optional activation function (default = ReLU)</summary>\r\n    activationFn?: IActivationFunction;\r\n}\r\n\r\n/// <summary>\r\n/// Type guard to check whether an object is an IMlpNeuron\r\n/// </summary>\r\nexport function isMlpNeuron(obj: unknown): obj is IMlpNeuron {\r\n    return typeof obj === \"object\" && obj !== null && \"bias\" in obj && \"activationFn\" in obj;\r\n}\r\n\r\n/// <summary>\r\n/// Represents a synapse (connection) in an MLP network.\r\n/// </summary>\r\nexport interface IMlpSynapse extends ISynapse {}\r\n\r\n/// <summary>\r\n/// A complete MLP graph composed of neurons and synapses.\r\n/// </summary>\r\nexport interface IMlpGraph extends IGraph<IMlpNeuron, IMlpSynapse> {}\r\n"]}

package/dist/neuralnetwork/ann/mlp/mlp.neuron.d.ts

@@ -0,0 +1,10 @@
+import { ICartesian } from "../../../geometry";
+import { IOlink } from "../../../graph";
+import { Nullable } from "../../../types";
+import { Neuron } from "../../nn.neuron";
+import { IActivationFunction, IMlpNeuron } from "./mlp.interfaces";
+export declare class MlpNeuron extends Neuron implements IMlpNeuron {
+    bias: number;
+    activationFn?: IActivationFunction | undefined;
+    constructor(bias?: number, activation?: IActivationFunction, onsc?: Nullable<IOlink[]>, opsc?: Nullable<IOlink[]>, position?: ICartesian);
+}

package/dist/neuralnetwork/ann/mlp/mlp.neuron.js

@@ -0,0 +1,20 @@
+import { __decorate, __metadata } from "tslib";
+import { cloneable } from "../../../graph";
+import { Neuron } from "../../nn.neuron";
+export class MlpNeuron extends Neuron {
+    constructor(bias = 0, activation, onsc = null, opsc = null, position) {
+        super(onsc, opsc, position);
+        this.bias = 0;
+        this.bias = bias;
+        this.activationFn = activation;
+    }
+}
+__decorate([
+    cloneable,
+    __metadata("design:type", Number)
+], MlpNeuron.prototype, "bias", void 0);
+__decorate([
+    cloneable,
+    __metadata("design:type", Object)
+], MlpNeuron.prototype, "activationFn", void 0);
+//# sourceMappingURL=mlp.neuron.js.map

package/dist/neuralnetwork/ann/mlp/mlp.neuron.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"mlp.neuron.js","sourceRoot":"","sources":["../../../../src/neuralnetwork/ann/mlp/mlp.neuron.ts"],"names":[],"mappings":";AACA,OAAO,EAAE,SAAS,EAAU,MAAM,gBAAgB,CAAC;AAEnD,OAAO,EAAE,MAAM,EAAE,MAAM,iBAAiB,CAAC;AAGzC,MAAM,OAAO,SAAU,SAAQ,MAAM;IAIjC,YAAmB,OAAe,CAAC,EAAE,UAAgC,EAAE,OAA2B,IAAI,EAAE,OAA2B,IAAI,EAAE,QAAqB;QAC1J,KAAK,CAAC,IAAI,EAAE,IAAI,EAAE,QAAQ,CAAC,CAAC;QAJd,SAAI,GAAW,CAAC,CAAC;QAK/B,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC;QACjB,IAAI,CAAC,YAAY,GAAG,UAAU,CAAC;IACnC,CAAC;CACJ;AARqB;IAAjB,SAAS;;uCAAyB;AACxB;IAAV,SAAS;;+CAAgD","sourcesContent":["import { ICartesian } from \"../../../geometry\";\r\nimport { cloneable, IOlink } from \"../../../graph\";\r\nimport { Nullable } from \"../../../types\";\r\nimport { Neuron } from \"../../nn.neuron\";\r\nimport { IActivationFunction, IMlpNeuron } from \"./mlp.interfaces\";\r\n\r\nexport class MlpNeuron extends Neuron implements IMlpNeuron {\r\n    @cloneable public bias: number = 0;\r\n    @cloneable activationFn?: IActivationFunction | undefined;\r\n\r\n    public constructor(bias: number = 0, activation?: IActivationFunction, onsc: Nullable<IOlink[]> = null, opsc: Nullable<IOlink[]> = null, position?: ICartesian) {\r\n        super(onsc, opsc, position);\r\n        this.bias = bias;\r\n        this.activationFn = activation;\r\n    }\r\n}\r\n"]}

package/dist/neuralnetwork/ann/mlp/mlp.runtime.utils.d.ts

@@ -0,0 +1,5 @@
+import { IMlpNeuron, IMlpSynapse } from "./mlp.interfaces";
+export declare class MLPRuntimeUtils {
+    static resetInferenceContext(neuron: IMlpNeuron): void;
+    static resetBackpropContext(item: IMlpNeuron | IMlpSynapse): void;
+}

package/dist/neuralnetwork/ann/mlp/mlp.runtime.utils.js

@@ -0,0 +1,48 @@
+import { isMlpNeuron } from "./mlp.interfaces";
+export class MLPRuntimeUtils {
+    static resetInferenceContext(neuron) {
+        if (!neuron.bag) {
+            const numInputs = neuron.opsc()?.length ?? 0;
+            neuron.bag = { sum: 0, activation: 0, remainingInputs: numInputs, totalInputs: numInputs };
+        }
+        else {
+            const bag = neuron.bag;
+            bag.sum = 0;
+            bag.activation = 0;
+            bag.remainingInputs = bag.totalInputs;
+        }
+    }
+    static resetBackpropContext(item) {
+        if (isMlpNeuron(item)) {
+            if (!item.bag) {
+                item.bag = { error: 0, gradient: 0 };
+            }
+            else {
+                const bag = item.bag;
+                bag.error = 0;
+                bag.gradient = 0;
+            }
+            return;
+        }
+        if (!item.bag) {
+            item.bag = {
+                gradient: 0,
+                velocity: 0,
+                m: 0,
+                v: 0,
+                prelookedWeight: 0,
+                weightDelta: 0,
+            };
+        }
+        else {
+            const bag = item.bag;
+            bag.gradient = 0;
+            bag.velocity = 0;
+            bag.m = 0;
+            bag.v = 0;
+            bag.prelookedWeight = 0;
+            bag.weightDelta = 0;
+        }
+    }
+}
+//# sourceMappingURL=mlp.runtime.utils.js.map

package/dist/neuralnetwork/ann/mlp/mlp.runtime.utils.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"mlp.runtime.utils.js","sourceRoot":"","sources":["../../../../src/neuralnetwork/ann/mlp/mlp.runtime.utils.ts"],"names":[],"mappings":"AAAA,OAAO,EAAoD,WAAW,EAAE,MAAM,kBAAkB,CAAC;AAGjG,MAAM,OAAO,eAAe;IACjB,MAAM,CAAC,qBAAqB,CAAC,MAAkB;QAClD,IAAI,CAAC,MAAM,CAAC,GAAG,EAAE,CAAC;YACd,MAAM,SAAS,GAAG,MAAM,CAAC,IAAI,EAAe,EAAE,MAAM,IAAI,CAAC,CAAC;YAC1D,MAAM,CAAC,GAAG,GAAG,EAAE,GAAG,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE,eAAe,EAAE,SAAS,EAAE,WAAW,EAAE,SAAS,EAAE,CAAC;QAC/F,CAAC;aAAM,CAAC;YACJ,MAAM,GAAG,GAAG,MAAM,CAAC,GAA8B,CAAC;YAClD,GAAG,CAAC,GAAG,GAAG,CAAC,CAAC;YACZ,GAAG,CAAC,UAAU,GAAG,CAAC,CAAC;YACnB,GAAG,CAAC,eAAe,GAAG,GAAG,CAAC,WAAW,CAAC;QAC1C,CAAC;IACL,CAAC;IAEM,MAAM,CAAC,oBAAoB,CAAC,IAA8B;QAC7D,IAAI,WAAW,CAAC,IAAI,CAAC,EAAE,CAAC;YACpB,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC;gBACZ,IAAI,CAAC,GAAG,GAAG,EAAE,KAAK,EAAE,CAAC,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC;YACzC,CAAC;iBAAM,CAAC;gBACJ,MAAM,GAAG,GAAG,IAAI,CAAC,GAA6B,CAAC;gBAC/C,GAAG,CAAC,KAAK,GAAG,CAAC,CAAC;gBACd,GAAG,CAAC,QAAQ,GAAG,CAAC,CAAC;YACrB,CAAC;YACD,OAAO;QACX,CAAC;QACD,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC;YACZ,IAAI,CAAC,GAAG,GAAG;gBACP,QAAQ,EAAE,CAAC;gBACX,QAAQ,EAAE,CAAC;gBACX,CAAC,EAAE,CAAC;gBACJ,CAAC,EAAE,CAAC;gBACJ,eAAe,EAAE,CAAC;gBAClB,WAAW,EAAE,CAAC;aACjB,CAAC;QACN,CAAC;aAAM,CAAC;YACJ,MAAM,GAAG,GAAG,IAAI,CAAC,GAA8B,CAAC;YAChD,GAAG,CAAC,QAAQ,GAAG,CAAC,CAAC;YACjB,GAAG,CAAC,QAAQ,GAAG,CAAC,CAAC;YACjB,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC;YACV,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC;YACV,GAAG,CAAC,eAAe,GAAG,CAAC,CAAC;YACxB,GAAG,CAAC,WAAW,GAAG,CAAC,CAAC;QACxB,CAAC;IACL,CAAC;CACJ","sourcesContent":["import { IMlpNeuron, IMlpSynapse, IInferenceNeuronContext, isMlpNeuron } from \"./mlp.interfaces\";\r\nimport { IBackpropNeuronContext, IBackpropSynapseContext } from \"./training\";\r\n\r\nexport class MLPRuntimeUtils {\r\n    public static resetInferenceContext(neuron: IMlpNeuron): void {\r\n        if (!neuron.bag) {\r\n            const numInputs = neuron.opsc<IMlpSynapse>()?.length ?? 0;\r\n            neuron.bag = { sum: 0, activation: 0, remainingInputs: numInputs, totalInputs: numInputs };\r\n        } else {\r\n            const bag = neuron.bag as IInferenceNeuronContext;\r\n            bag.sum = 0;\r\n            bag.activation = 0;\r\n            bag.remainingInputs = bag.totalInputs;\r\n        }\r\n    }\r\n\r\n    public static resetBackpropContext(item: IMlpNeuron | IMlpSynapse): void {\r\n        if (isMlpNeuron(item)) {\r\n            if (!item.bag) {\r\n                item.bag = { error: 0, gradient: 0 };\r\n            } else {\r\n                const bag = item.bag as IBackpropNeuronContext;\r\n                bag.error = 0;\r\n                bag.gradient = 0;\r\n            }\r\n            return;\r\n        }\r\n        if (!item.bag) {\r\n            item.bag = {\r\n                gradient: 0,\r\n                velocity: 0,\r\n                m: 0,\r\n                v: 0,\r\n                prelookedWeight: 0,\r\n                weightDelta: 0,\r\n            };\r\n        } else {\r\n            const bag = item.bag as IBackpropSynapseContext;\r\n            bag.gradient = 0;\r\n            bag.velocity = 0;\r\n            bag.m = 0;\r\n            bag.v = 0;\r\n            bag.prelookedWeight = 0;\r\n            bag.weightDelta = 0;\r\n        }\r\n    }\r\n}\r\n"]}

package/dist/neuralnetwork/ann/mlp/mlp.synapse.d.ts

@@ -0,0 +1,6 @@
+import { INode } from "../../../graph";
+import { Synapse } from "../../nn.synapse";
+import { IMlpSynapse } from "./mlp.interfaces";
+export declare class MlpSynapse extends Synapse implements IMlpSynapse {
+    constructor(oini: INode, ofin: INode);
+}

package/dist/neuralnetwork/ann/mlp/mlp.synapse.js

@@ -0,0 +1,7 @@
+import { Synapse } from "../../nn.synapse";
+export class MlpSynapse extends Synapse {
+    constructor(oini, ofin) {
+        super(oini, ofin);
+    }
+}
+//# sourceMappingURL=mlp.synapse.js.map

package/dist/neuralnetwork/ann/mlp/mlp.synapse.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"mlp.synapse.js","sourceRoot":"","sources":["../../../../src/neuralnetwork/ann/mlp/mlp.synapse.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,OAAO,EAAE,MAAM,kBAAkB,CAAC;AAG3C,MAAM,OAAO,UAAW,SAAQ,OAAO;IACnC,YAAmB,IAAW,EAAE,IAAW;QACvC,KAAK,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;IACtB,CAAC;CACJ","sourcesContent":["import { INode } from \"../../../graph\";\r\nimport { Synapse } from \"../../nn.synapse\";\r\nimport { IMlpSynapse } from \"./mlp.interfaces\";\r\n\r\nexport class MlpSynapse extends Synapse implements IMlpSynapse {\r\n    public constructor(oini: INode, ofin: INode) {\r\n        super(oini, ofin);\r\n    }\r\n}\r\n"]}

package/dist/neuralnetwork/ann/mlp/training/index.d.ts

@@ -0,0 +1,4 @@
+export * from "./mlp.optimizers";
+export * from "./mlp.loss";
+export * from "./mlp.training.interfaces";
+export * from "./mlp.training";

package/dist/neuralnetwork/ann/mlp/training/index.js

@@ -0,0 +1,5 @@
+export * from "./mlp.optimizers";
+export * from "./mlp.loss";
+export * from "./mlp.training.interfaces";
+export * from "./mlp.training";
+//# sourceMappingURL=index.js.map

package/dist/neuralnetwork/ann/mlp/training/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../../../src/neuralnetwork/ann/mlp/training/index.ts"],"names":[],"mappings":"AAAA,cAAc,kBAAkB,CAAC;AACjC,cAAc,YAAY,CAAC;AAC3B,cAAc,2BAA2B,CAAC;AAC1C,cAAc,gBAAgB,CAAC","sourcesContent":["export * from \"./mlp.optimizers\";\r\nexport * from \"./mlp.loss\";\r\nexport * from \"./mlp.training.interfaces\";\r\nexport * from \"./mlp.training\";\r\n"]}

package/dist/neuralnetwork/ann/mlp/training/mlp.loss.d.ts

@@ -0,0 +1,10 @@
+export declare const LossFunctions: {
+    MSE: {
+        loss: (o: number, y: number) => number;
+        dLoss: (o: number, y: number) => number;
+    };
+    CrossEntropy: {
+        loss: (o: number, y: number) => number;
+        dLoss: (o: number, y: number) => number;
+    };
+};

package/dist/neuralnetwork/ann/mlp/training/mlp.loss.js

@@ -0,0 +1,11 @@
+export const LossFunctions = {
+    MSE: {
+        loss: (o, y) => 0.5 * Math.pow(o - y, 2),
+        dLoss: (o, y) => o - y,
+    },
+    CrossEntropy: {
+        loss: (o, y) => -y * Math.log(o + 1e-8) - (1 - y) * Math.log(1 - o + 1e-8),
+        dLoss: (o, y) => (o - y) / ((o + 1e-8) * (1 - o + 1e-8)),
+    },
+};
+//# sourceMappingURL=mlp.loss.js.map

package/dist/neuralnetwork/ann/mlp/training/mlp.loss.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"mlp.loss.js","sourceRoot":"","sources":["../../../../../src/neuralnetwork/ann/mlp/training/mlp.loss.ts"],"names":[],"mappings":"AAEA,MAAM,CAAC,MAAM,aAAa,GAAG;IACzB,GAAG,EAAE;QACD,IAAI,EAAE,CAAC,CAAS,EAAE,CAAS,EAAE,EAAE,CAAC,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;QACxD,KAAK,EAAE,CAAC,CAAS,EAAE,CAAS,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC;KACzC;IACD,YAAY,EAAE;QACV,IAAI,EAAE,CAAC,CAAS,EAAE,CAAS,EAAE,EAAE,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,GAAG,IAAI,CAAC;QAC1F,KAAK,EAAE,CAAC,CAAS,EAAE,CAAS,EAAE,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,GAAG,IAAI,CAAC,CAAC;KAC3E;CACoC,CAAC","sourcesContent":["import { ILossFunction } from \"./mlp.training.interfaces\";\r\n\r\nexport const LossFunctions = {\r\n    MSE: {\r\n        loss: (o: number, y: number) => 0.5 * Math.pow(o - y, 2),\r\n        dLoss: (o: number, y: number) => o - y,\r\n    },\r\n    CrossEntropy: {\r\n        loss: (o: number, y: number) => -y * Math.log(o + 1e-8) - (1 - y) * Math.log(1 - o + 1e-8),\r\n        dLoss: (o: number, y: number) => (o - y) / ((o + 1e-8) * (1 - o + 1e-8)),\r\n    },\r\n} satisfies Record<string, ILossFunction>;\r\n"]}

package/dist/neuralnetwork/ann/mlp/training/mlp.optimizers.d.ts

@@ -0,0 +1,7 @@
+import { IOptimizer } from "./mlp.training.interfaces";
+export declare class Optimizers {
+    static SGD: () => IOptimizer;
+    static MomentumSGD: (momentum: number) => IOptimizer;
+    static NAG: (momentum?: number) => IOptimizer;
+    static Adam: (beta1?: number, beta2?: number, epsilon?: number) => IOptimizer;
+}

package/dist/neuralnetwork/ann/mlp/training/mlp.optimizers.js

@@ -0,0 +1,59 @@
+/// <summary>
+/// Stochastic Gradient Descent (SGD) optimizer
+export class Optimizers {
+}
+/// </summary>
+Optimizers.SGD = () => ({
+    apply(synapse, lr, gradient, ctx) {
+        const bag = (synapse.bag ??= { gradient: gradient });
+        bag.gradient = gradient;
+        bag.weightDelta = -lr * gradient;
+        synapse.weight += bag.weightDelta;
+    },
+});
+/// <summary>
+/// SGD with Momentum optimizer
+/// </summary>
+Optimizers.MomentumSGD = (momentum) => ({
+    apply(synapse, lr, gradient, ctx) {
+        const bag = (synapse.bag ??= { gradient: gradient });
+        bag.gradient = gradient;
+        bag.velocity ??= 0;
+        bag.velocity = momentum * bag.velocity - lr * gradient;
+        bag.weightDelta = bag.velocity;
+        synapse.weight += bag.weightDelta;
+    },
+});
+/// <summary>
+/// Nesterov Accelerated Gradient (NAG) optimizer
+/// </summary>
+Optimizers.NAG = (momentum = 0.9) => ({
+    apply(synapse, lr, gradient, ctx) {
+        const bag = (synapse.bag ??= { gradient: gradient });
+        bag.gradient = gradient;
+        bag.velocity ??= 0;
+        bag.velocity = momentum * bag.velocity - lr * gradient;
+        bag.weightDelta = bag.velocity;
+        synapse.weight += bag.weightDelta;
+        bag.prelookedWeight = undefined;
+    },
+});
+/// <summary>
+/// Adam optimizer with bias correction
+/// </summary>
+Optimizers.Adam = (beta1 = 0.9, beta2 = 0.999, epsilon = 1e-8) => ({
+    apply(synapse, lr, gradient, ctx) {
+        const bag = (synapse.bag ??= { gradient: gradient });
+        const t = ctx.iteration + 1;
+        bag.gradient = gradient;
+        bag.m ??= 0;
+        bag.v ??= 0;
+        bag.m = beta1 * bag.m + (1 - beta1) * gradient;
+        bag.v = beta2 * bag.v + (1 - beta2) * gradient * gradient;
+        const mHat = bag.m / (1 - Math.pow(beta1, t));
+        const vHat = bag.v / (1 - Math.pow(beta2, t));
+        bag.weightDelta = (-lr * mHat) / (Math.sqrt(vHat) + epsilon);
+        synapse.weight += bag.weightDelta;
+    },
+});
+//# sourceMappingURL=mlp.optimizers.js.map