UNPKG

22.4 kBJavaScriptView Raw
1"use strict";
2/**
3 * @license
4 * Copyright 2018 Google LLC. All Rights Reserved.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at
8 *
9 * http://www.apache.org/licenses/LICENSE-2.0
10 *
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
16 * =============================================================================
17 */
18var __awaiter = (this && this.__awaiter) || function (thisArg, _arguments, P, generator) {
19 return new (P || (P = Promise))(function (resolve, reject) {
20 function fulfilled(value) { try { step(generator.next(value)); } catch (e) { reject(e); } }
21 function rejected(value) { try { step(generator["throw"](value)); } catch (e) { reject(e); } }
22 function step(result) { result.done ? resolve(result.value) : new P(function (resolve) { resolve(result.value); }).then(fulfilled, rejected); }
23 step((generator = generator.apply(thisArg, _arguments || [])).next());
24 });
25};
26var __generator = (this && this.__generator) || function (thisArg, body) {
27 var _ = { label: 0, sent: function() { if (t[0] & 1) throw t[1]; return t[1]; }, trys: [], ops: [] }, f, y, t, g;
28 return g = { next: verb(0), "throw": verb(1), "return": verb(2) }, typeof Symbol === "function" && (g[Symbol.iterator] = function() { return this; }), g;
29 function verb(n) { return function (v) { return step([n, v]); }; }
30 function step(op) {
31 if (f) throw new TypeError("Generator is already executing.");
32 while (_) try {
33 if (f = 1, y && (t = op[0] & 2 ? y["return"] : op[0] ? y["throw"] || ((t = y["return"]) && t.call(y), 0) : y.next) && !(t = t.call(y, op[1])).done) return t;
34 if (y = 0, t) op = [op[0] & 2, t.value];
35 switch (op[0]) {
36 case 0: case 1: t = op; break;
37 case 4: _.label++; return { value: op[1], done: false };
38 case 5: _.label++; y = op[1]; op = [0]; continue;
39 case 7: op = _.ops.pop(); _.trys.pop(); continue;
40 default:
41 if (!(t = _.trys, t = t.length > 0 && t[t.length - 1]) && (op[0] === 6 || op[0] === 2)) { _ = 0; continue; }
42 if (op[0] === 3 && (!t || (op[1] > t[0] && op[1] < t[3]))) { _.label = op[1]; break; }
43 if (op[0] === 6 && _.label < t[1]) { _.label = t[1]; t = op; break; }
44 if (t && _.label < t[2]) { _.label = t[2]; _.ops.push(op); break; }
45 if (t[2]) _.ops.pop();
46 _.trys.pop(); continue;
47 }
48 op = body.call(thisArg, _);
49 } catch (e) { op = [6, e]; y = 0; } finally { f = t = 0; }
50 if (op[0] & 5) throw op[1]; return { value: op[0] ? op[1] : void 0, done: true };
51 }
52};
53var _this = this;
54Object.defineProperty(exports, "__esModule", { value: true });
55var tf = require("@tensorflow/tfjs");
56var callbacks_1 = require("./callbacks");
57describe('progbarLogger', function () {
58 // Fake progbar class written for testing.
59 var FakeProgbar = /** @class */ (function () {
60 function FakeProgbar(specs, config) {
61 this.specs = specs;
62 this.config = config;
63 this.tickConfigs = [];
64 }
65 FakeProgbar.prototype.tick = function (tickConfig) {
66 this.tickConfigs.push(tickConfig);
67 };
68 return FakeProgbar;
69 }());
70 var originalStderrColumns;
71 beforeEach(function () {
72 // In some CI environments, process.stderr.columns has a null value.
73 originalStderrColumns = process.stderr.columns;
74 process.stderr.columns = 100;
75 });
76 afterEach(function () {
77 process.stderr.columns = originalStderrColumns;
78 });
79 it('Model.fit with loss, no metric, no validation, verobse = 1', function () { return __awaiter(_this, void 0, void 0, function () {
80 var fakeProgbars, consoleMessages, model, numSamples, epochs, batchSize, xs, ys, _i, fakeProgbars_1, fakeProgbar, tickConfigs, i;
81 return __generator(this, function (_a) {
82 switch (_a.label) {
83 case 0:
84 fakeProgbars = [];
85 spyOn(callbacks_1.progressBarHelper, 'ProgressBar')
86 .and.callFake(function (specs, config) {
87 var fakeProgbar = new FakeProgbar(specs, config);
88 fakeProgbars.push(fakeProgbar);
89 return fakeProgbar;
90 });
91 consoleMessages = [];
92 spyOn(callbacks_1.progressBarHelper, 'log').and.callFake(function (message) {
93 consoleMessages.push(message);
94 });
95 model = tf.sequential();
96 model.add(tf.layers.dense({ units: 10, inputShape: [8], activation: 'relu' }));
97 model.add(tf.layers.dense({ units: 1 }));
98 model.compile({ loss: 'meanSquaredError', optimizer: 'sgd' });
99 numSamples = 14;
100 epochs = 3;
101 batchSize = 8;
102 xs = tf.randomNormal([numSamples, 8]);
103 ys = tf.randomNormal([numSamples, 1]);
104 return [4 /*yield*/, model.fit(xs, ys, { epochs: epochs, batchSize: batchSize, verbose: 1 })];
105 case 1:
106 _a.sent();
107 // A progbar object is created for each epoch.
108 expect(fakeProgbars.length).toEqual(3);
109 for (_i = 0, fakeProgbars_1 = fakeProgbars; _i < fakeProgbars_1.length; _i++) {
110 fakeProgbar = fakeProgbars_1[_i];
111 tickConfigs = fakeProgbar.tickConfigs;
112 // There are ceil(14 / 8) = 2 batchs per epoch. There should be 1 tick
113 // for epoch batch, plus a tick at the end of the epoch.
114 expect(tickConfigs.length).toEqual(3);
115 for (i = 0; i < 2; ++i) {
116 expect(Object.keys(tickConfigs[i])).toEqual([
117 'placeholderForLossesAndMetrics'
118 ]);
119 expect(tickConfigs[i]['placeholderForLossesAndMetrics'])
120 .toMatch(/^loss=.*/);
121 }
122 expect(tickConfigs[2]).toEqual({ placeholderForLossesAndMetrics: '' });
123 }
124 expect(consoleMessages.length).toEqual(6);
125 expect(consoleMessages[0]).toEqual('Epoch 1 / 3');
126 expect(consoleMessages[1]).toMatch(/.*ms .*us\/step - loss=.*/);
127 expect(consoleMessages[2]).toEqual('Epoch 2 / 3');
128 expect(consoleMessages[3]).toMatch(/.*ms .*us\/step - loss=.*/);
129 expect(consoleMessages[4]).toEqual('Epoch 3 / 3');
130 expect(consoleMessages[5]).toMatch(/.*ms .*us\/step - loss=.*/);
131 return [2 /*return*/];
132 }
133 });
134 }); });
135 it('Model.fit with loss, metric and validation, verbose = 2', function () { return __awaiter(_this, void 0, void 0, function () {
136 var fakeProgbars, consoleMessages, model, numSamples, epochs, batchSize, validationSplit, xs, ys, _i, fakeProgbars_2, fakeProgbar, tickConfigs, i;
137 return __generator(this, function (_a) {
138 switch (_a.label) {
139 case 0:
140 fakeProgbars = [];
141 spyOn(callbacks_1.progressBarHelper, 'ProgressBar')
142 .and.callFake(function (specs, config) {
143 var fakeProgbar = new FakeProgbar(specs, config);
144 fakeProgbars.push(fakeProgbar);
145 return fakeProgbar;
146 });
147 consoleMessages = [];
148 spyOn(callbacks_1.progressBarHelper, 'log').and.callFake(function (message) {
149 consoleMessages.push(message);
150 });
151 model = tf.sequential();
152 model.add(tf.layers.dense({ units: 10, inputShape: [8], activation: 'relu' }));
153 model.add(tf.layers.dense({ units: 1 }));
154 model.compile({ loss: 'meanSquaredError', optimizer: 'sgd', metrics: ['acc'] });
155 numSamples = 40;
156 epochs = 2;
157 batchSize = 8;
158 validationSplit = 0.15;
159 xs = tf.randomNormal([numSamples, 8]);
160 ys = tf.randomNormal([numSamples, 1]);
161 return [4 /*yield*/, model.fit(xs, ys, { epochs: epochs, batchSize: batchSize, validationSplit: validationSplit, verbose: 2 })];
162 case 1:
163 _a.sent();
164 // A progbar object is created for each epoch.
165 expect(fakeProgbars.length).toEqual(2);
166 for (_i = 0, fakeProgbars_2 = fakeProgbars; _i < fakeProgbars_2.length; _i++) {
167 fakeProgbar = fakeProgbars_2[_i];
168 tickConfigs = fakeProgbar.tickConfigs;
169 // There are 5 batchs per epoch. There should be 1 tick for epoch batch,
170 // plus a tick at the end of the epoch.
171 expect(tickConfigs.length).toEqual(6);
172 for (i = 0; i < 5; ++i) {
173 expect(Object.keys(tickConfigs[i])).toEqual([
174 'placeholderForLossesAndMetrics'
175 ]);
176 expect(tickConfigs[i]['placeholderForLossesAndMetrics'])
177 .toMatch(/^acc=.* loss=.*/);
178 }
179 expect(tickConfigs[5]).toEqual({ placeholderForLossesAndMetrics: '' });
180 }
181 expect(consoleMessages.length).toEqual(4);
182 expect(consoleMessages[0]).toEqual('Epoch 1 / 2');
183 expect(consoleMessages[1])
184 .toMatch(/.*ms .*us\/step - acc=.* loss=.* val_acc=.* val_loss=.*/);
185 expect(consoleMessages[2]).toEqual('Epoch 2 / 2');
186 expect(consoleMessages[3])
187 .toMatch(/.*ms .*us\/step - acc=.* loss=.* val_acc=.* val_loss=.*/);
188 return [2 /*return*/];
189 }
190 });
191 }); });
192 it('Model.fit does not create ProgbarLogger if verbose is 0', function () { return __awaiter(_this, void 0, void 0, function () {
193 var fakeProgbars, consoleMessages, model, numSamples, epochs, batchSize, validationSplit, xs, ys;
194 return __generator(this, function (_a) {
195 switch (_a.label) {
196 case 0:
197 fakeProgbars = [];
198 spyOn(callbacks_1.progressBarHelper, 'ProgressBar')
199 .and.callFake(function (specs, config) {
200 var fakeProgbar = new FakeProgbar(specs, config);
201 fakeProgbars.push(fakeProgbar);
202 return fakeProgbar;
203 });
204 consoleMessages = [];
205 spyOn(callbacks_1.progressBarHelper, 'log').and.callFake(function (message) {
206 consoleMessages.push(message);
207 });
208 model = tf.sequential();
209 model.add(tf.layers.dense({ units: 10, inputShape: [8], activation: 'relu' }));
210 model.add(tf.layers.dense({ units: 1 }));
211 model.compile({ loss: 'meanSquaredError', optimizer: 'sgd', metrics: ['acc'] });
212 numSamples = 40;
213 epochs = 2;
214 batchSize = 8;
215 validationSplit = 0.15;
216 xs = tf.randomNormal([numSamples, 8]);
217 ys = tf.randomNormal([numSamples, 1]);
218 return [4 /*yield*/, model.fit(xs, ys, { epochs: epochs, batchSize: batchSize, validationSplit: validationSplit, verbose: 0 })];
219 case 1:
220 _a.sent();
221 expect(fakeProgbars.length).toEqual(0);
222 return [2 /*return*/];
223 }
224 });
225 }); });
226 it('Model.fitDataset: batchesPerEpoch specified, verbose = 1', function () { return __awaiter(_this, void 0, void 0, function () {
227 var fakeProgbars, consoleMessages, epochs, xDataset, yDataset, dataset, model;
228 return __generator(this, function (_a) {
229 switch (_a.label) {
230 case 0:
231 fakeProgbars = [];
232 spyOn(callbacks_1.progressBarHelper, 'ProgressBar')
233 .and.callFake(function (specs, config) {
234 var fakeProgbar = new FakeProgbar(specs, config);
235 fakeProgbars.push(fakeProgbar);
236 return fakeProgbar;
237 });
238 consoleMessages = [];
239 spyOn(callbacks_1.progressBarHelper, 'log').and.callFake(function (message) {
240 consoleMessages.push(message);
241 });
242 epochs = 2;
243 xDataset = tf.data.array([[1, 2], [3, 4], [5, 6], [7, 8]])
244 .map(function (x) { return tf.tensor2d(x, [1, 2]); });
245 yDataset = tf.data.array([[1], [2], [3], [4]]).map(function (y) { return tf.tensor2d(y, [1, 1]); });
246 dataset = tf.data.zip({ xs: xDataset, ys: yDataset }).repeat(epochs);
247 model = tf.sequential();
248 model.add(tf.layers.dense({ units: 1, inputShape: [2] }));
249 model.compile({ loss: 'meanSquaredError', optimizer: 'sgd' });
250 return [4 /*yield*/, model.fitDataset(dataset, { batchesPerEpoch: 4, epochs: epochs, verbose: 1 })];
251 case 1:
252 _a.sent();
253 expect(consoleMessages.length).toEqual(4);
254 expect(consoleMessages[0]).toEqual('Epoch 1 / 2');
255 expect(consoleMessages[1]).toMatch(/.*ms .*us\/step - loss=.*/);
256 expect(consoleMessages[2]).toEqual('Epoch 2 / 2');
257 expect(consoleMessages[3]).toMatch(/.*ms .*us\/step - loss=.*/);
258 return [2 /*return*/];
259 }
260 });
261 }); });
262 it('Model.fitDataset: batchesPerEpoch unavailable, verbose = 1', function () { return __awaiter(_this, void 0, void 0, function () {
263 var fakeProgbars, consoleMessages, epochs, xDataset, yDataset, dataset, model;
264 return __generator(this, function (_a) {
265 switch (_a.label) {
266 case 0:
267 fakeProgbars = [];
268 spyOn(callbacks_1.progressBarHelper, 'ProgressBar')
269 .and.callFake(function (specs, config) {
270 var fakeProgbar = new FakeProgbar(specs, config);
271 fakeProgbars.push(fakeProgbar);
272 return fakeProgbar;
273 });
274 consoleMessages = [];
275 spyOn(callbacks_1.progressBarHelper, 'log').and.callFake(function (message) {
276 consoleMessages.push(message);
277 });
278 epochs = 2;
279 xDataset = tf.data.array([[1, 2], [3, 4], [5, 6], [7, 8]])
280 .map(function (x) { return tf.tensor2d(x, [1, 2]); });
281 yDataset = tf.data.array([[1], [2], [3], [4]]).map(function (y) { return tf.tensor2d(y, [1, 1]); });
282 dataset = tf.data.zip({ xs: xDataset, ys: yDataset }).repeat(epochs);
283 model = tf.sequential();
284 model.add(tf.layers.dense({ units: 1, inputShape: [2] }));
285 model.compile({ loss: 'meanSquaredError', optimizer: 'sgd' });
286 // `batchesPerEpoch` is not specified. Instead, `fitDataset()` relies on
287 // the `done` field being `true` to terminate the epoch(s).
288 return [4 /*yield*/, model.fitDataset(dataset, { epochs: epochs, verbose: 1 })];
289 case 1:
290 // `batchesPerEpoch` is not specified. Instead, `fitDataset()` relies on
291 // the `done` field being `true` to terminate the epoch(s).
292 _a.sent();
293 expect(consoleMessages.length).toEqual(4);
294 expect(consoleMessages[0]).toEqual('Epoch 1 / 2');
295 expect(consoleMessages[1]).toMatch(/.*ms .*us\/step - loss=.*/);
296 expect(consoleMessages[2]).toEqual('Epoch 2 / 2');
297 expect(consoleMessages[3]).toMatch(/.*ms .*us\/step - loss=.*/);
298 return [2 /*return*/];
299 }
300 });
301 }); });
302 it('Model.fitDataset: verbose = 0 leads to no logging', function () { return __awaiter(_this, void 0, void 0, function () {
303 var fakeProgbars, consoleMessages, xDataset, yDataset, dataset, model, history;
304 return __generator(this, function (_a) {
305 switch (_a.label) {
306 case 0:
307 fakeProgbars = [];
308 spyOn(callbacks_1.progressBarHelper, 'ProgressBar')
309 .and.callFake(function (specs, config) {
310 var fakeProgbar = new FakeProgbar(specs, config);
311 fakeProgbars.push(fakeProgbar);
312 return fakeProgbar;
313 });
314 consoleMessages = [];
315 spyOn(callbacks_1.progressBarHelper, 'log').and.callFake(function (message) {
316 consoleMessages.push(message);
317 });
318 xDataset = tf.data.array([[1, 2], [3, 4], [5, 6], [7, 8]])
319 .map(function (x) { return tf.tensor2d(x, [1, 2]); });
320 yDataset = tf.data.array([[1], [2], [3], [4]]).map(function (y) { return tf.tensor2d(y, [1, 1]); });
321 dataset = tf.data.zip({ xs: xDataset, ys: yDataset });
322 model = tf.sequential();
323 model.add(tf.layers.dense({ units: 1, inputShape: [2] }));
324 model.compile({ loss: 'meanSquaredError', optimizer: 'sgd' });
325 return [4 /*yield*/, model.fitDataset(dataset, { epochs: 1, verbose: 0 })];
326 case 1:
327 history = _a.sent();
328 expect(history.history.loss.length).toEqual(1);
329 expect(consoleMessages.length)
330 .toEqual(0); // No logging should have happened.
331 return [2 /*return*/];
332 }
333 });
334 }); });
335});
336describe('getSuccinctNumberDisplay', function () {
337 it('Not finite', function () {
338 expect(callbacks_1.getSuccinctNumberDisplay(Infinity)).toEqual('Infinity');
339 expect(callbacks_1.getSuccinctNumberDisplay(-Infinity)).toEqual('-Infinity');
340 expect(callbacks_1.getSuccinctNumberDisplay(NaN)).toEqual('NaN');
341 });
342 it('zero', function () {
343 expect(callbacks_1.getSuccinctNumberDisplay(0)).toEqual('0.00');
344 });
345 it('Finite and positive', function () {
346 expect(callbacks_1.getSuccinctNumberDisplay(300)).toEqual('300.00');
347 expect(callbacks_1.getSuccinctNumberDisplay(30)).toEqual('30.00');
348 expect(callbacks_1.getSuccinctNumberDisplay(1)).toEqual('1.00');
349 expect(callbacks_1.getSuccinctNumberDisplay(1e-2)).toEqual('0.0100');
350 expect(callbacks_1.getSuccinctNumberDisplay(1e-3)).toEqual('1.00e-3');
351 expect(callbacks_1.getSuccinctNumberDisplay(4e-3)).toEqual('4.00e-3');
352 expect(callbacks_1.getSuccinctNumberDisplay(1e-6)).toEqual('1.00e-6');
353 });
354 it('Finite and negative', function () {
355 expect(callbacks_1.getSuccinctNumberDisplay(-300)).toEqual('-300.00');
356 expect(callbacks_1.getSuccinctNumberDisplay(-30)).toEqual('-30.00');
357 expect(callbacks_1.getSuccinctNumberDisplay(-1)).toEqual('-1.00');
358 expect(callbacks_1.getSuccinctNumberDisplay(-1e-2)).toEqual('-0.0100');
359 expect(callbacks_1.getSuccinctNumberDisplay(-1e-3)).toEqual('-1.00e-3');
360 expect(callbacks_1.getSuccinctNumberDisplay(-4e-3)).toEqual('-4.00e-3');
361 expect(callbacks_1.getSuccinctNumberDisplay(-1e-6)).toEqual('-1.00e-6');
362 });
363});
364describe('getDisplayDecimalPlaces', function () {
365 it('Not finite', function () {
366 expect(callbacks_1.getDisplayDecimalPlaces(Infinity)).toEqual(2);
367 expect(callbacks_1.getDisplayDecimalPlaces(-Infinity)).toEqual(2);
368 expect(callbacks_1.getDisplayDecimalPlaces(NaN)).toEqual(2);
369 });
370 it('zero', function () {
371 expect(callbacks_1.getDisplayDecimalPlaces(0)).toEqual(2);
372 });
373 it('Finite and positive', function () {
374 expect(callbacks_1.getDisplayDecimalPlaces(300)).toEqual(2);
375 expect(callbacks_1.getDisplayDecimalPlaces(30)).toEqual(2);
376 expect(callbacks_1.getDisplayDecimalPlaces(1)).toEqual(2);
377 expect(callbacks_1.getDisplayDecimalPlaces(1e-2)).toEqual(4);
378 expect(callbacks_1.getDisplayDecimalPlaces(1e-3)).toEqual(5);
379 expect(callbacks_1.getDisplayDecimalPlaces(4e-3)).toEqual(5);
380 expect(callbacks_1.getDisplayDecimalPlaces(1e-6)).toEqual(8);
381 });
382 it('Finite and negative', function () {
383 expect(callbacks_1.getDisplayDecimalPlaces(-300)).toEqual(2);
384 expect(callbacks_1.getDisplayDecimalPlaces(-30)).toEqual(2);
385 expect(callbacks_1.getDisplayDecimalPlaces(-1)).toEqual(2);
386 expect(callbacks_1.getDisplayDecimalPlaces(-1e-2)).toEqual(4);
387 expect(callbacks_1.getDisplayDecimalPlaces(-1e-3)).toEqual(5);
388 expect(callbacks_1.getDisplayDecimalPlaces(-4e-3)).toEqual(5);
389 expect(callbacks_1.getDisplayDecimalPlaces(-1e-6)).toEqual(8);
390 });
391});