Try to add an additional input to the neural network, but it's not working.

.tern-project

@@ -1,7 +1,11 @@
 {
   "ecmaVersion": 8,
-  "libs": [],
-  "loadEagerly": [],
+  "libs": [
+    "browser"
+  ],
+  "loadEagerly": [
+    "server/**.mjs"
+  ],
   "dontLoad": [],
   "plugins": {
     "doc_comment": true

common/Math.mjs

@@ -6,6 +6,13 @@ function normalise(value, { min : input_min, max: input_max }, { min : output_mi
 	) + output_min
 }
 
+/**
+ * Clamps a value to fit within the specified bounds.
+ * @param  {number} value The number to clamp.
+ * @param  {number} min   The minimum value the number can be.
+ * @param  {number} max   The maximum value the number can be.
+ * @return {number}       The clamped number.
+ */
 function clamp(value, min, max) {
 	if(value > max) return max;
 	if(value < min) return min;

server/Repos.SQLite/GatewayRepo.mjs

@@ -36,6 +36,10 @@ class GatewayRepo {
 		return count > 0;
 	}
 	
+	get_by_id(id) {
+		return this.db.prepare(`SELECT * FROM gateways WHERE id = :id`).get({ id });
+	}
+	
 	/**
 	 * Iterates over all the gateways in the table.
 	 * TODO: Use Symbol.iterator here?

server/Repos.SQLite/ReadingRepo.mjs

@@ -39,6 +39,10 @@ class ReadingRepo {
 		return count > 0;
 	}
 	
+	iterate_unreceived() {
+		return this.db.prepare(`SELECT * FROM readings WHERE data_rate IS NULL;`).iterate();
+	}
+	
 	iterate() {
 		return this.db.prepare(`SELECT * FROM readings`).iterate();
 	}

server/settings.default.toml

@@ -66,6 +66,8 @@ rssi_max = 0
 batch_size = 32
 # The number of epochs to train for.
 epochs = 5
+# The percentage, between 0 and 1, of data that should be set aside for validation.
+validation_split = 0.1
 
 # The directory to output trained AIs to, relative to the repository root.
 output_directory = "app/ais/"

server/train-ai/AITrainer.mjs

@@ -6,7 +6,8 @@ import fs from 'fs';
 import tf from '@tensorflow/tfjs-node-gpu';
 
 class AITrainer {
-	constructor({ settings, log, root_dir, GatewayRepo, DatasetFetcher }) {
+	constructor({ ansi, settings, log, root_dir, GatewayRepo, DatasetFetcher }) {
+		this.a = ansi;
 		this.settings = settings;
 		this.root_dir = root_dir;
 		this.l = log;
@@ -17,9 +18,9 @@ class AITrainer {
 	generate_model() {
 		let model = tf.sequential();
 		model.add(tf.layers.dense({
-			units: 256, // 256 nodes
+			units: 64, // 64 nodes
 			activation: "sigmoid", // Sigmoid activation function
-			inputShape: [2], // 2 inputs - lat and long
+			inputShape: [3], // 3 inputs - lat, long, and distance from gateway
 		}))
 		model.add(tf.layers.dense({
 			units: 1, // 1 output value - RSSI
@@ -82,30 +83,36 @@ class AITrainer {
 	 * @return	{Promise}	A promise that resolves when training and serialisation is complete.
 	 */
 	async train_gateway(gateway_id, destination_filename) {
+		this.l.log(`${this.a.fgreen}${this.a.hicol}Training AI for gateway ${gateway_id}${this.a.reset}`);
 		let model = this.generate_model();
 		
-		// TODO: Add samples here for locations that the gateway does NOT cover too
-		let dataset_input = tf.data.generator(
-			this.dataset_fetcher.fetch_input.bind(this.dataset_fetcher, gateway_id)
-		);
-		let dataset_output = tf.data.generator(
-			this.dataset_fetcher.fetch_output.bind(this.dataset_fetcher, gateway_id)
-		);
-		
-		let dataset = tf.data.zip({
-				xs: dataset_input,
-				ys: dataset_output
-			}).shuffle(this.settings.ai.batch_size * 4)
-			.batch(this.settings.ai.batch_size);
-		
+		// let dataset_input = tf.data.generator(
+		// 	this.dataset_fetcher.fetch_input.bind(this.dataset_fetcher, gateway_id)
+		// );
+		// let dataset_output = tf.data.generator(
+		// 	this.dataset_fetcher.fetch_output.bind(this.dataset_fetcher, gateway_id)
+		// );
+		// 
+		// let dataset = tf.data.zip({
+		// 		xs: dataset_input,
+		// 		ys: dataset_output
+		// 	}).shuffle(this.settings.ai.batch_size * 4)
+		// 	.batch(this.settings.ai.batch_size);
+		// 
+		let datasets = this.dataset_fetcher.fetch_all(gateway_id);
 		
-		let result = await model.fitDataset(dataset, {
-			epochs: this.settings.ai.epochs,
-			batchSize:  this.settings.ai.batch_size
-		});
+		let result = await model.fit(
+			tf.tensor(datasets.input),
+			tf.tensor(datasets.output),
+			{
+				epochs: this.settings.ai.epochs,
+				batchSize:  this.settings.ai.batch_size,
+				validationSplit: this.settings.ai.validation_split
+			}
+		);
 		
 		await model.save(`file://${destination_filename}`);
-		console.log(result);
+		// console.log(result);
 		
 		return true;
 	}

server/train-ai/DatasetFetcher.mjs

@@ -1,26 +1,78 @@
 "use strict";
 
+import haversine from 'haversine-distance';
+
 import { normalise, clamp } from '../../common/Math.mjs';
 
 class DatasetFetcher {
-	constructor({ settings, RSSIRepo }) {
+	constructor({ settings, GatewayRepo, RSSIRepo, ReadingRepo }) {
 		this.settings = settings;
+		this.repo_gateway = GatewayRepo;
 		this.repo_rssi = RSSIRepo;
+		this.repo_reading = ReadingRepo;
 	}
 	
-	*fetch_input(gateway_id) {
+	normalise_latlng(lat, lng) {
+		return [
+			normalise(lat,
+				{ min: -90, max: +90 },
+				{ min: 0, max: 1 }
+			),
+			normalise(lng,
+				{ min: -180, max: +180 },
+				{ min: 0, max: 1 }
+			)
+		];
+	}
+	
+	fetch_all(gateway_id) {
+		let gateway_location = this.repo_gateway.get_by_id(gateway_id);
+		
+		let result = {
+			input: [],
+			output: []
+		};
 		for(let rssi of this.repo_rssi.iterate_gateway(gateway_id)) {
-			yield [
-				normalise(rssi.latitude,
-					{ min: -90, max: +90 },
+			let next_input = this.normalise_latlng(rssi.latitude, rssi.longitude);
+			let distance_from_gateway = haversine(gateway_location, rssi);
+			
+			next_input.push(clamp(
+				normalise(distance_from_gateway,
+					{ min: 0, max: 20000 },
 					{ min: 0, max: 1 }
 				),
-				normalise(rssi.longitude,
-					{ min: -180, max: +180 },
+			0, 1))
+			
+			result.input.push(next_input);
+			
+			console.log(`Distance from gateway: ${haversine(gateway_location, rssi)}m`);
+			
+			result.output.push([
+				clamp(normalise(rssi.rssi,
+					{ min: this.settings.ai.rssi_min, max: this.settings.ai.rssi_max },
 					{ min: 0, max: 1 }
-				)
-			];
+				), 0, 1)
+			]);
 		}
+		
+		for(let reading of this.repo_reading.iterate_unreceived()) {
+			result.input.push(this.normalise_latlng(
+				reading.latitude,
+				reading.longitude
+			));
+			
+			result.output.push([ 0 ]);
+		}
+		
+		return result;
+	}
+	
+	*fetch_input(gateway_id) {
+		for(let rssi of this.repo_rssi.iterate_gateway(gateway_id))
+			yield this.normalise_latlng(rssi.latitude, rssi.longitude);
+		
+		for(let rssi of this.repo_reading.iterate_unreceived())
+			yield this.normalise_latlng(rssi.latitude, rssi.longitude);
 	}
 	
 	*fetch_output(gateway_id) {
@@ -32,6 +84,10 @@ class DatasetFetcher {
 				), 0, 1)
 			];
 		}
+		// Yield 0 for every unreceived message, since we want to train it to predict a *terrible* signal where the gateway is not
+		for(let rssi of this.repo_reading.iterate_unreceived()) {
+			yield [ 0 ];
+		}
 	}
 }