Tests for the neuralNetwork addData function.

src/NeuralNetwork/index.test.js

@@ -0,0 +1,160 @@
+import { asyncLoadImage, randomImageData } from '../utils/testingUtils';
+import neuralNetwork from './index';
+
+describe('Adding data to Neural Network', () => {
+  let nn;
+
+  const lastData = () =>
+    nn.neuralNetworkData.data.raw[nn.neuralNetworkData.data.raw.length - 1];
+
+  const warnSpy = jest.spyOn(console, 'warn');
+
+  describe('classification', () => {
+    describe('using property names for inputs and outputs', () => {
+      beforeAll(async () => {
+        return new Promise((resolve) => {
+          nn = neuralNetwork({
+            inputs: ['avg_temperature', 'humidity'],
+            outputs: ['rained'],
+            task: 'classification'
+          }, () => resolve());
+        });
+      });
+
+      it('creates a model', () => {
+        expect(nn.options.inputs).toHaveLength(2);
+        expect(nn.options.outputs).toHaveLength(1);
+        expect(nn.options.outputs[0]).toBe('rained');
+      });
+
+      it('can add data from a dictionary object', () => {
+        nn.addData({ "avg_temperature": 20, "humidity": 0.2 }, { "rained": "no" });
+        const added = lastData();
+        expect(added.xs.avg_temperature).toBe(20);
+        expect(added.xs.humidity).toBe(0.2);
+        expect(added.ys.rained).toBe('no');
+      });
+
+      // Currently fails
+      it.skip('ignores excess properties', () => {
+        // TODO: why do xs and ys need to be passed separately in this case?
+        const row = { "avg_temperature": 30, "humidity": 0.9, "rained": "yes" };
+        nn.addData(row, row);
+        const added = lastData();
+        expect(added.xs.humidity).toBe(0.9);
+        expect(added.xs.rained).toBeUndefined();
+        expect(added.ys.rained).toBe('yes');
+        expect(added.ys.humidity).toBeUndefined();
+      });
+
+      // Currently fails
+      it.skip('errors on missing properties', () => {
+        expect(() => {
+          nn.addData({}, { "rained": "yes" });
+        }).toThrow(); // TODO: check for a friendly message
+      });
+
+      it('can add data as an array', () => {
+        nn.addData([25, 0.3], ["no"]);
+        const added = lastData();
+        expect(added.xs.avg_temperature).toBe(25);
+        expect(added.xs.humidity).toBe(0.3);
+        expect(added.ys.rained).toBe('no');
+      });
+
+      // Currently fails
+      it.skip('errors on too few inputs', () => {
+        expect(() => {
+          nn.addData([25], ["no"]);
+        }).toThrow(); // TODO: check for a friendly message
+      });
+
+      // Currently fails
+      it.skip('warns on too many inputs', () => {
+        expect(() => {
+          nn.addData([1, 2, 3, 4, 5], ["no"]);
+        }).not.toThrow();
+        expect(warnSpy).toHaveBeenCalled();
+      });
+    });
+
+    describe('using numbers for inputs and outputs', () => {
+      beforeEach(async () => {
+        return new Promise((resolve) => {
+          nn = neuralNetwork({
+            inputs: 2,
+            outputs: 2,
+            task: 'classification'
+          }, () => resolve());
+        });
+      });
+
+      it('creates a model', () => {
+        expect(nn.options.inputs).toBe(2);
+        expect(nn.options.outputs).toBe(2);
+      });
+
+      it('can add data from a dictionary object', async () => {
+        nn.addData({ "avg_temperature": 25, "humidity": 0.3 }, { "rained": "no" });
+        const added = lastData();
+        expect(added.xs.avg_temperature).toBe(25);
+        expect(added.xs.humidity).toBe(0.3);
+        expect(added.ys.rained).toBe('no');
+        nn.addData({ "avg_temperature": 30, "humidity": 0.9 },{ "rained": "yes" });
+        await nn.train();
+        const guess = await nn.classify({ "avg_temperature": 20, "humidity": 0.4 });
+        expect(guess[0].label).toBe('no');
+        expect(guess[1].label).toBe('yes');
+      });
+
+      it('can add data as an array', async () => {
+        nn.addData([25, 0.3], ["no"]);
+        const added = lastData();
+        expect(added.xs['0']).toBe(25);
+        expect(added.xs['1']).toBe(0.3);
+        expect(added.ys['0']).toBe('no');
+        nn.addData([30, 0.9], ["yes"]);
+        await nn.train();
+        const guess = await nn.classify([20, 0.4]);
+        expect(guess[0].label).toBe('no');
+        expect(guess[1].label).toBe('yes');
+      });
+    });
+  });
+
+  // All currently fail
+  describe.skip('image classification', () => {
+    beforeAll(async () => {
+      nn = neuralNetwork({
+        task: 'imageClassification',
+        inputs: [32, 32, 3],
+        // outputs: ['label']
+        outputs: 2
+        // outputs: ['cat', 'dog']
+      });
+      await nn.ready;
+    });
+
+    it('can add HTML images', () => {
+      const dataSpy = jest.spyOn(nn.neuralNetworkData, 'addData');
+      const image = asyncLoadImage('https://cdn.jsdelivr.net/gh/ml5js/ml5-library@main/assets/bird.jpg');
+      nn.addData({ image: image }, { label: 'cat' });
+      expect(dataSpy).toHaveBeenCalledTimes(1);
+      expect(lastData().xs.image.length).toBe(32 * 32 * 3);
+      // TODO
+    });
+
+    it('can add untyped pixel arrays', () => {
+      nn.addData({ image: new Array(32 * 32 * 3).fill(0) }, { label: 'cat' });
+      nn.addData(new Array(32 * 32 * 3).fill(0), 'cat');
+      // TODO
+    });
+
+    it('can add typed pixel arrays', () => {
+      const imageData = randomImageData(32, 32);
+      nn.addData({ image: imageData.data }, { label: 'cat' });
+      nn.addData(imageData.data, 'cat');
+      // TODO
+    });
+  });
+});