@tensorflow/tfjs-core
Version:
Hardware-accelerated JavaScript library for machine intelligence
1,001 lines • 67.7 kB
JavaScript
"use strict";
/**
* @license
* Copyright 2019 Google LLC. All Rights Reserved.
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
* =============================================================================
*/
var __awaiter = (this && this.__awaiter) || function (thisArg, _arguments, P, generator) {
return new (P || (P = Promise))(function (resolve, reject) {
function fulfilled(value) { try { step(generator.next(value)); } catch (e) { reject(e); } }
function rejected(value) { try { step(generator["throw"](value)); } catch (e) { reject(e); } }
function step(result) { result.done ? resolve(result.value) : new P(function (resolve) { resolve(result.value); }).then(fulfilled, rejected); }
step((generator = generator.apply(thisArg, _arguments || [])).next());
});
};
var __generator = (this && this.__generator) || function (thisArg, body) {
var _ = { label: 0, sent: function() { if (t[0] & 1) throw t[1]; return t[1]; }, trys: [], ops: [] }, f, y, t, g;
return g = { next: verb(0), "throw": verb(1), "return": verb(2) }, typeof Symbol === "function" && (g[Symbol.iterator] = function() { return this; }), g;
function verb(n) { return function (v) { return step([n, v]); }; }
function step(op) {
if (f) throw new TypeError("Generator is already executing.");
while (_) try {
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;
if (y = 0, t) op = [op[0] & 2, t.value];
switch (op[0]) {
case 0: case 1: t = op; break;
case 4: _.label++; return { value: op[1], done: false };
case 5: _.label++; y = op[1]; op = [0]; continue;
case 7: op = _.ops.pop(); _.trys.pop(); continue;
default:
if (!(t = _.trys, t = t.length > 0 && t[t.length - 1]) && (op[0] === 6 || op[0] === 2)) { _ = 0; continue; }
if (op[0] === 3 && (!t || (op[1] > t[0] && op[1] < t[3]))) { _.label = op[1]; break; }
if (op[0] === 6 && _.label < t[1]) { _.label = t[1]; t = op; break; }
if (t && _.label < t[2]) { _.label = t[2]; _.ops.push(op); break; }
if (t[2]) _.ops.pop();
_.trys.pop(); continue;
}
op = body.call(thisArg, _);
} catch (e) { op = [6, e]; y = 0; } finally { f = t = 0; }
if (op[0] & 5) throw op[1]; return { value: op[0] ? op[1] : void 0, done: true };
}
};
var _this = this;
Object.defineProperty(exports, "__esModule", { value: true });
var tf = require("../index");
var jasmine_util_1 = require("../jasmine_util");
var test_util_1 = require("../test_util");
jasmine_util_1.describeWithFlags('fused matmul', jasmine_util_1.ALL_ENVS, function () {
it('A x B', function () { return __awaiter(_this, void 0, void 0, function () {
var a, b, c, _a;
return __generator(this, function (_b) {
switch (_b.label) {
case 0:
a = tf.tensor2d([1, 2, 3, 4, 5, 6], [2, 3]);
b = tf.tensor2d([0, 1, -3, 2, 2, 1], [3, 2]);
c = tf.fused.matMul({ a: a, b: b });
expect(c.shape).toEqual([2, 2]);
_a = test_util_1.expectArraysClose;
return [4 /*yield*/, c.data()];
case 1:
_a.apply(void 0, [_b.sent(), [0, 8, -3, 20]]);
return [2 /*return*/];
}
});
}); });
it('A x B with relu', function () { return __awaiter(_this, void 0, void 0, function () {
var a, b, transposeA, transposeB, c, _a;
return __generator(this, function (_b) {
switch (_b.label) {
case 0:
a = tf.tensor2d([1, 2, 3, 4, 5, 6], [2, 3]);
b = tf.tensor2d([0, 1, -3, 2, 2, 1], [3, 2]);
transposeA = false;
transposeB = false;
c = tf.fused.matMul({ a: a, b: b, transposeA: transposeA, transposeB: transposeB, bias: null, activation: 'relu' });
expect(c.shape).toEqual([2, 2]);
_a = test_util_1.expectArraysClose;
return [4 /*yield*/, c.data()];
case 1:
_a.apply(void 0, [_b.sent(), [0, 8, 0, 20]]);
return [2 /*return*/];
}
});
}); });
it('A x B with elu', function () { return __awaiter(_this, void 0, void 0, function () {
var a, b, transposeA, transposeB, c, _a;
return __generator(this, function (_b) {
switch (_b.label) {
case 0:
a = tf.tensor2d([1, 2, 3, 4, 5, 6], [2, 3]);
b = tf.tensor2d([0, 1, -3, 2, 2, 1], [3, 2]);
transposeA = false;
transposeB = false;
c = tf.fused.matMul({ a: a, b: b, transposeA: transposeA, transposeB: transposeB, bias: null, activation: 'elu' });
expect(c.shape).toEqual([2, 2]);
_a = test_util_1.expectArraysClose;
return [4 /*yield*/, c.data()];
case 1:
_a.apply(void 0, [_b.sent(), [0, 8, -0.9502, 20]]);
return [2 /*return*/];
}
});
}); });
it('A x B with relu6', function () { return __awaiter(_this, void 0, void 0, function () {
var a, b, transposeA, transposeB, c, _a;
return __generator(this, function (_b) {
switch (_b.label) {
case 0:
a = tf.tensor2d([1, 2, 3, 4, 5, 6], [2, 3]);
b = tf.tensor2d([0, 1, -3, 2, 2, 1], [3, 2]);
transposeA = false;
transposeB = false;
c = tf.fused.matMul({ a: a, b: b, transposeA: transposeA, transposeB: transposeB, bias: null, activation: 'relu6' });
expect(c.shape).toEqual([2, 2]);
_a = test_util_1.expectArraysClose;
return [4 /*yield*/, c.data()];
case 1:
_a.apply(void 0, [_b.sent(), [0, 6, 0, 6]]);
return [2 /*return*/];
}
});
}); });
it('A x B with prelu', function () { return __awaiter(_this, void 0, void 0, function () {
var a, b, alpha, transposeA, transposeB, c, _a;
return __generator(this, function (_b) {
switch (_b.label) {
case 0:
a = tf.tensor2d([1, 2, 3, 4, 5, 6], [2, 3]);
b = tf.tensor2d([0, 1, -3, 2, 2, 1], [3, 2]);
alpha = tf.tensor2d([0.5, 0.5], [1, 2]);
transposeA = false;
transposeB = false;
c = tf.fused.matMul({
a: a,
b: b,
transposeA: transposeA,
transposeB: transposeB,
bias: null,
activation: 'prelu',
preluActivationWeights: alpha
});
expect(c.shape).toEqual([2, 2]);
_a = test_util_1.expectArraysClose;
return [4 /*yield*/, c.data()];
case 1:
_a.apply(void 0, [_b.sent(), [0, 8, -1.5, 20]]);
return [2 /*return*/];
}
});
}); });
it('A x B with relu transpose', function () { return __awaiter(_this, void 0, void 0, function () {
var a, b, transposeA, transposeB, c, _a;
return __generator(this, function (_b) {
switch (_b.label) {
case 0:
a = tf.tensor2d([1, 2, 3, 4, 5, 6], [2, 3]);
b = tf.tensor2d([0, 1, -3, 2, 2, 1], [2, 3]);
transposeA = false;
transposeB = true;
c = tf.fused.matMul({ a: a, b: b, transposeA: transposeA, transposeB: transposeB, bias: null, activation: 'relu' });
expect(c.shape).toEqual([2, 2]);
_a = test_util_1.expectArraysClose;
return [4 /*yield*/, c.data()];
case 1:
_a.apply(void 0, [_b.sent(), [0, 9, 0, 24]]);
return [2 /*return*/];
}
});
}); });
it('A x B with relu and bias', function () { return __awaiter(_this, void 0, void 0, function () {
var a, b, c, transposeA, transposeB, d, _a;
return __generator(this, function (_b) {
switch (_b.label) {
case 0:
a = tf.tensor2d([1, 2, 3, 4, 5, 6], [2, 3]);
b = tf.tensor2d([0, 1, -3, 2, 2, 1], [3, 2]);
c = tf.tensor2d([1, 1, 1, 1], [2, 2]);
transposeA = false;
transposeB = false;
d = tf.fused.matMul({ a: a, b: b, transposeA: transposeA, transposeB: transposeB, bias: c, activation: 'relu' });
expect(d.shape).toEqual([2, 2]);
_a = test_util_1.expectArraysClose;
return [4 /*yield*/, d.data()];
case 1:
_a.apply(void 0, [_b.sent(), [1, 9, 0, 21]]);
return [2 /*return*/];
}
});
}); });
it('A x B with relu and broadcasted bias', function () { return __awaiter(_this, void 0, void 0, function () {
var a, b, c, act, transposeA, transposeB, d, _a;
return __generator(this, function (_b) {
switch (_b.label) {
case 0:
a = tf.tensor2d([1, 2, 3, 4, 5, 6], [2, 3]);
b = tf.tensor2d([0, 1, -3, 2, 2, 1], [3, 2]);
c = tf.tensor1d([1, 1]);
act = 'relu';
transposeA = false;
transposeB = false;
d = tf.fused.matMul({ a: a, b: b, transposeA: transposeA, transposeB: transposeB, bias: c, activation: act });
expect(d.shape).toEqual([2, 2]);
_a = test_util_1.expectArraysClose;
return [4 /*yield*/, d.data()];
case 1:
_a.apply(void 0, [_b.sent(), [1, 9, 0, 21]]);
return [2 /*return*/];
}
});
}); });
it('A x B with elu and broadcasted bias', function () { return __awaiter(_this, void 0, void 0, function () {
var a, b, c, act, transposeA, transposeB, d, _a;
return __generator(this, function (_b) {
switch (_b.label) {
case 0:
a = tf.tensor2d([1, 2, 3, 4, 5, 6], [2, 3]);
b = tf.tensor2d([0, 1, -3, 2, 2, 1], [3, 2]);
c = tf.tensor1d([1, 1]);
act = 'elu';
transposeA = false;
transposeB = false;
d = tf.fused.matMul({ a: a, b: b, transposeA: transposeA, transposeB: transposeB, bias: c, activation: act });
expect(d.shape).toEqual([2, 2]);
_a = test_util_1.expectArraysClose;
return [4 /*yield*/, d.data()];
case 1:
_a.apply(void 0, [_b.sent(), [1, 9, -0.8647, 21]]);
return [2 /*return*/];
}
});
}); });
it('A x B with relu and broadcasted bias different rank', function () { return __awaiter(_this, void 0, void 0, function () {
var a, b, c, act, transposeA, transposeB, d, _a;
return __generator(this, function (_b) {
switch (_b.label) {
case 0:
a = tf.tensor3d([0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11], [2, 2, 3]);
b = tf.tensor3d([0, 1, -3, 2, 2, 1, 0, 1, -3, 2, 2, 1], [2, 3, 2]);
c = tf.tensor2d([1, 2], [1, 2]);
act = 'relu';
transposeA = false;
transposeB = false;
d = tf.fused.matMul({ a: a, b: b, transposeA: transposeA, transposeB: transposeB, bias: c, activation: act });
expect(d.shape).toEqual([2, 2, 2]);
_a = test_util_1.expectArraysClose;
return [4 /*yield*/, d.data()];
case 1:
_a.apply(void 0, [_b.sent(), [2, 6, 0, 18, 0, 30, 0, 42]]);
return [2 /*return*/];
}
});
}); });
it('A x B with bias only', function () { return __awaiter(_this, void 0, void 0, function () {
var a, b, c, transposeA, transposeB, d, _a;
return __generator(this, function (_b) {
switch (_b.label) {
case 0:
a = tf.tensor2d([1, 2, 3, 4, 5, 6], [2, 3]);
b = tf.tensor2d([0, 1, -3, 2, 2, 1], [3, 2]);
c = tf.tensor2d([1, 1, 1, 1], [2, 2]);
transposeA = false;
transposeB = false;
d = tf.fused.matMul({ a: a, b: b, transposeA: transposeA, transposeB: transposeB, bias: c, activation: 'linear' });
expect(d.shape).toEqual([2, 2]);
_a = test_util_1.expectArraysClose;
return [4 /*yield*/, d.data()];
case 1:
_a.apply(void 0, [_b.sent(), [1, 9, -2, 21]]);
return [2 /*return*/];
}
});
}); });
it('A x B with relu gradient', function () { return __awaiter(_this, void 0, void 0, function () {
var a, b, dy, transposeA, transposeB, grads, fusedGrads, _a, da, db, _b, fusedDa, fusedDb, _c, _d, _e, _f;
return __generator(this, function (_g) {
switch (_g.label) {
case 0:
a = tf.tensor2d([1, 2, 3, 10, 20, -30], [2, 3]);
b = tf.tensor2d([2, 3, 4, -1, 2, 3], [3, 2]);
dy = tf.tensor2d([1, 10, 20, 30], [2, 2]);
transposeA = false;
transposeB = false;
grads = tf.grads(function (a, b) {
var prod = tf.matMul(a, b, transposeA, transposeB);
return tf.relu(prod);
});
fusedGrads = tf.grads(function (a, b) {
return tf.fused.matMul({ a: a, b: b, transposeA: transposeA, transposeB: transposeB, bias: null, activation: 'relu' });
});
_a = grads([a, b], dy), da = _a[0], db = _a[1];
_b = fusedGrads([a, b], dy), fusedDa = _b[0], fusedDb = _b[1];
_c = test_util_1.expectArraysClose;
return [4 /*yield*/, da.array()];
case 1:
_d = [_g.sent()];
return [4 /*yield*/, fusedDa.array()];
case 2:
_c.apply(void 0, _d.concat([_g.sent()]));
_e = test_util_1.expectArraysClose;
return [4 /*yield*/, db.data()];
case 3:
_f = [_g.sent()];
return [4 /*yield*/, fusedDb.array()];
case 4:
_e.apply(void 0, _f.concat([_g.sent()]));
return [2 /*return*/];
}
});
}); });
it('gradient with clones A x B with relu', function () {
var a = tf.tensor2d([1, 2, 3, 10, 20, -30], [2, 3]);
var b = tf.tensor2d([2, 3, 4, -1, 2, 3], [3, 2]);
var dy = tf.tensor2d([1, 10, 20, 30], [2, 2]);
var transposeA = false;
var transposeB = false;
var fusedGrads = tf.grads(function (a, b) {
return tf.fused
.matMul({
a: a.clone(),
b: b.clone(),
transposeA: transposeA,
transposeB: transposeB,
bias: null,
activation: 'relu'
})
.clone();
});
var _a = fusedGrads([a, b], dy), fusedDa = _a[0], fusedDb = _a[1];
expect(fusedDa.shape).toEqual(a.shape);
expect(fusedDb.shape).toEqual(b.shape);
});
it('A x B with relu bias gradient', function () { return __awaiter(_this, void 0, void 0, function () {
var a, b, c, transposeA, transposeB, dy, grads, fusedGrads, _a, da, db, dc, _b, fusedDa, fusedDb, fusedDc, _c, _d, _e, _f, _g, _h;
return __generator(this, function (_j) {
switch (_j.label) {
case 0:
a = tf.tensor2d([1, 2, 3, 10, 20, -30], [2, 3]);
b = tf.tensor2d([2, 3, 4, -1, 2, 3], [3, 2]);
c = tf.tensor2d([1, 1, 1, 1], [2, 2]);
transposeA = false;
transposeB = false;
dy = tf.tensor2d([1, 10, 20, 30], [2, 2]);
grads = tf.grads(function (a, b, c) {
var prod = tf.matMul(a, b, transposeA, transposeB);
var sum = tf.add(prod, c);
return tf.relu(sum);
});
fusedGrads = tf.grads(function (a, b, c) {
return tf.fused.matMul({ a: a, b: b, transposeA: transposeA, transposeB: transposeB, bias: c, activation: 'relu' });
});
_a = grads([a, b, c], dy), da = _a[0], db = _a[1], dc = _a[2];
_b = fusedGrads([a, b, c], dy), fusedDa = _b[0], fusedDb = _b[1], fusedDc = _b[2];
_c = test_util_1.expectArraysClose;
return [4 /*yield*/, da.array()];
case 1:
_d = [_j.sent()];
return [4 /*yield*/, fusedDa.array()];
case 2:
_c.apply(void 0, _d.concat([_j.sent()]));
_e = test_util_1.expectArraysClose;
return [4 /*yield*/, db.array()];
case 3:
_f = [_j.sent()];
return [4 /*yield*/, fusedDb.array()];
case 4:
_e.apply(void 0, _f.concat([_j.sent()]));
_g = test_util_1.expectArraysClose;
return [4 /*yield*/, dc.array()];
case 5:
_h = [_j.sent()];
return [4 /*yield*/, fusedDc.array()];
case 6:
_g.apply(void 0, _h.concat([_j.sent()]));
return [2 /*return*/];
}
});
}); });
it('A x B with relu bias gradient transpose', function () { return __awaiter(_this, void 0, void 0, function () {
var a, b, c, transposeA, transposeB, dy, grads, fusedGrads, _a, da, db, dc, _b, fusedDa, fusedDb, fusedDc, _c, _d, _e, _f, _g, _h;
return __generator(this, function (_j) {
switch (_j.label) {
case 0:
a = tf.tensor2d([1, 2, 3, 10, 20, -30], [3, 2]);
b = tf.tensor2d([2, 3, 4, -1, 2, 3], [3, 2]);
c = tf.tensor2d([1, 1, 1, 1], [2, 2]);
transposeA = true;
transposeB = false;
dy = tf.tensor2d([1, 10, 20, 30], [2, 2]);
grads = tf.grads(function (a, b, c) {
var prod = tf.matMul(a, b, transposeA, transposeB);
var sum = tf.add(prod, c);
return tf.relu(sum);
});
fusedGrads = tf.grads(function (a, b, c) {
return tf.fused.matMul({ a: a, b: b, transposeA: transposeA, transposeB: transposeB, bias: c, activation: 'relu' });
});
_a = grads([a, b, c], dy), da = _a[0], db = _a[1], dc = _a[2];
_b = fusedGrads([a, b, c], dy), fusedDa = _b[0], fusedDb = _b[1], fusedDc = _b[2];
_c = test_util_1.expectArraysClose;
return [4 /*yield*/, da.array()];
case 1:
_d = [_j.sent()];
return [4 /*yield*/, fusedDa.array()];
case 2:
_c.apply(void 0, _d.concat([_j.sent()]));
_e = test_util_1.expectArraysClose;
return [4 /*yield*/, db.array()];
case 3:
_f = [_j.sent()];
return [4 /*yield*/, fusedDb.array()];
case 4:
_e.apply(void 0, _f.concat([_j.sent()]));
_g = test_util_1.expectArraysClose;
return [4 /*yield*/, dc.array()];
case 5:
_h = [_j.sent()];
return [4 /*yield*/, fusedDc.array()];
case 6:
_g.apply(void 0, _h.concat([_j.sent()]));
return [2 /*return*/];
}
});
}); });
it('A x B with relu and broadcasted bias gradient', function () { return __awaiter(_this, void 0, void 0, function () {
var a, b, c, transposeA, transposeB, dy, grads, fusedGrads, _a, da, db, dc, _b, fusedDa, fusedDb, fusedDc, _c, _d, _e, _f, _g, _h;
return __generator(this, function (_j) {
switch (_j.label) {
case 0:
a = tf.tensor2d([1, 2, 3, 10, 20, -30], [2, 3]);
b = tf.tensor2d([2, 3, 4, -1, 2, 3], [3, 2]);
c = tf.tensor2d([[1]]);
transposeA = false;
transposeB = false;
dy = tf.tensor2d([1, 10, 20, 30], [2, 2]);
grads = tf.grads(function (a, b, c) {
var prod = tf.matMul(a, b, transposeA, transposeB);
var sum = tf.add(prod, c);
return tf.relu(sum);
});
fusedGrads = tf.grads(function (a, b, c) {
return tf.fused.matMul({ a: a, b: b, transposeA: transposeA, transposeB: transposeB, bias: c, activation: 'relu' });
});
_a = grads([a, b, c], dy), da = _a[0], db = _a[1], dc = _a[2];
_b = fusedGrads([a, b, c], dy), fusedDa = _b[0], fusedDb = _b[1], fusedDc = _b[2];
_c = test_util_1.expectArraysClose;
return [4 /*yield*/, da.array()];
case 1:
_d = [_j.sent()];
return [4 /*yield*/, fusedDa.array()];
case 2:
_c.apply(void 0, _d.concat([_j.sent()]));
_e = test_util_1.expectArraysClose;
return [4 /*yield*/, db.array()];
case 3:
_f = [_j.sent()];
return [4 /*yield*/, fusedDb.array()];
case 4:
_e.apply(void 0, _f.concat([_j.sent()]));
_g = test_util_1.expectArraysClose;
return [4 /*yield*/, dc.array()];
case 5:
_h = [_j.sent()];
return [4 /*yield*/, fusedDc.array()];
case 6:
_g.apply(void 0, _h.concat([_j.sent()]));
return [2 /*return*/];
}
});
}); });
});
jasmine_util_1.describeWithFlags('fused depthwiseConv2D', jasmine_util_1.ALL_ENVS, function () {
it('basic', function () { return __awaiter(_this, void 0, void 0, function () {
var fSize, pad, strides, chMul, inDepth, x, w, result, expected, _a;
return __generator(this, function (_b) {
switch (_b.label) {
case 0:
fSize = 2;
pad = 'valid';
strides = 1;
chMul = 1;
inDepth = 1;
x = tf.tensor4d([
0.230664, 0.987388, 0.0685208, 0.419224, 0.887861, 0.731641,
0.0741907, 0.409265, 0.351377
], [1, 3, 3, inDepth]);
w = tf.tensor4d([-0.303873, -0.229223, 0.144333, 0.803373], [fSize, fSize, inDepth, chMul]);
result = tf.fused.depthwiseConv2d({ x: x, filter: w, strides: strides, pad: pad });
expect(result.shape).toEqual([1, 2, 2, 1]);
expected = [0.47737, 0.40018, 0.00859, -0.09615];
_a = test_util_1.expectArraysClose;
return [4 /*yield*/, result.data()];
case 1:
_a.apply(void 0, [_b.sent(), expected]);
return [2 /*return*/];
}
});
}); });
it('basic with relu', function () { return __awaiter(_this, void 0, void 0, function () {
var fSize, pad, strides, chMul, inDepth, x, w, result, expected, _a;
return __generator(this, function (_b) {
switch (_b.label) {
case 0:
fSize = 2;
pad = 'valid';
strides = 1;
chMul = 1;
inDepth = 1;
x = tf.tensor4d([
0.230664, 0.987388, 0.0685208, 0.419224, 0.887861, 0.731641,
0.0741907, 0.409265, 0.351377
], [1, 3, 3, inDepth]);
w = tf.tensor4d([-0.303873, -0.229223, 0.144333, 0.803373], [fSize, fSize, inDepth, chMul]);
result = tf.fused.depthwiseConv2d({ x: x, filter: w, strides: strides, pad: pad, activation: 'relu' });
expect(result.shape).toEqual([1, 2, 2, 1]);
expected = [0.47737, 0.40018, 0.00859, 0];
_a = test_util_1.expectArraysClose;
return [4 /*yield*/, result.data()];
case 1:
_a.apply(void 0, [_b.sent(), expected]);
return [2 /*return*/];
}
});
}); });
it('basic with broadcasted bias and relu', function () { return __awaiter(_this, void 0, void 0, function () {
var fSize, pad, strides, chMul, inDepth, x, w, result, expected, _a;
return __generator(this, function (_b) {
switch (_b.label) {
case 0:
fSize = 2;
pad = 'valid';
strides = 1;
chMul = 1;
inDepth = 1;
x = tf.tensor4d([
0.230664, 0.987388, 0.0685208, 0.419224, 0.887861, 0.731641,
0.0741907, 0.409265, 0.351377
], [1, 3, 3, inDepth]);
w = tf.tensor4d([-0.303873, -0.229223, 0.144333, 0.803373], [fSize, fSize, inDepth, chMul]);
result = tf.fused.depthwiseConv2d({ x: x, filter: w, strides: strides, pad: pad, bias: tf.scalar(1), activation: 'relu' });
expect(result.shape).toEqual([1, 2, 2, 1]);
expected = [1.47737, 1.40018, 1.00859, 0.90385];
_a = test_util_1.expectArraysClose;
return [4 /*yield*/, result.data()];
case 1:
_a.apply(void 0, [_b.sent(), expected]);
return [2 /*return*/];
}
});
}); });
it('prelu', function () { return __awaiter(_this, void 0, void 0, function () {
var fSize, pad, strides, chMul, inDepth, x, alpha, w, result, expected, _a;
return __generator(this, function (_b) {
switch (_b.label) {
case 0:
fSize = 3;
pad = 'valid';
strides = 1;
chMul = 1;
inDepth = 1;
x = tf.tensor4d([
0.149194, 0.089009, 0.654891, 0.083324, 0.537043, 0.644331, 0.563037,
0.211859, 0.633501, 0.186427, 0.777034, 0.50001, 0.607341, 0.95303,
0.696479, 0.050387, 0.62045, 0.728049, 0.028043, 0.437009, 0.712881,
0.741935, 0.974474, 0.621102, 0.171411
], [1, 5, 5, inDepth]);
alpha = tf.tensor4d([0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9], [1, 3, 3, 1]);
w = tf.tensor4d([
-0.125386, -0.975199, -0.640437, -0.281895, -0.990968, -0.347208,
-0.889702, -0.180695, -0.691992
], [fSize, fSize, inDepth, chMul]);
result = tf.fused.depthwiseConv2d({
x: x,
filter: w,
strides: strides,
pad: pad,
activation: 'prelu',
preluActivationWeights: alpha
});
expect(result.shape).toEqual([1, 3, 3, 1]);
expected = [
-0.25400, -0.50118, -0.73622, -0.94068, -1.2298, -1.84585, -2.3089,
-2.7499, -2.64077
];
_a = test_util_1.expectArraysClose;
return [4 /*yield*/, result.data()];
case 1:
_a.apply(void 0, [_b.sent(), expected]);
return [2 /*return*/];
}
});
}); });
it('gradient x=[2,3,3,1] f=[2,2,1,1] s=1 p=0', function () { return __awaiter(_this, void 0, void 0, function () {
var inputDepth, outputDepth, inputShape, filterSize, strides, pad, filterShape, filter, x, dy, grads, _a, dx, dfilter, _b, _c;
return __generator(this, function (_d) {
switch (_d.label) {
case 0:
inputDepth = 1;
outputDepth = 1;
inputShape = [2, 3, 3, inputDepth];
filterSize = 2;
strides = 1;
pad = 0;
filterShape = [filterSize, filterSize, inputDepth, outputDepth];
filter = tf.tensor4d([-1, 1, -2, 0.5], filterShape);
x = tf.tensor4d([1, 2, 3, 4, 5, 6, 7, 8, 9, 1, 2, 3, 4, 5, 6, 7, 8, 9], inputShape);
dy = tf.tensor4d([3, 1, 2, 0, 3, 1, 2, 0], [2, 2, 2, 1]);
grads = tf.grads(function (x, filter) {
return tf.fused.depthwiseConv2d({ x: x, filter: filter, strides: strides, pad: pad });
});
_a = grads([x, filter], dy), dx = _a[0], dfilter = _a[1];
expect(dx.shape).toEqual(x.shape);
_b = test_util_1.expectArraysClose;
return [4 /*yield*/, dx.data()];
case 1:
_b.apply(void 0, [_d.sent(),
[-3, 2, 1, -8, 1.5, 0.5, -4, 1, 0, -3, 2, 1, -8, 1.5, 0.5, -4, 1, 0]]);
expect(dfilter.shape).toEqual(filterShape);
_c = test_util_1.expectArraysClose;
return [4 /*yield*/, dfilter.data()];
case 2:
_c.apply(void 0, [_d.sent(), [26, 38, 62, 74]]);
return [2 /*return*/];
}
});
}); });
it('gradient x=[2,3,3,1] f=[2,2,1,1] s=1 p=0 with bias', function () { return __awaiter(_this, void 0, void 0, function () {
var inputDepth, outputDepth, inputShape, filterSize, strides, pad, filterShape, filter, bias, x, dy, fusedGrads, _a, dxFused, dfilterFused, dbiasFused, grads, _b, dx, dfilter, dbias, _c, _d, _e, _f, _g, _h;
return __generator(this, function (_j) {
switch (_j.label) {
case 0:
inputDepth = 1;
outputDepth = 1;
inputShape = [2, 3, 3, inputDepth];
filterSize = 2;
strides = 1;
pad = 0;
filterShape = [filterSize, filterSize, inputDepth, outputDepth];
filter = tf.tensor4d([-1, 1, -2, 0.5], filterShape);
bias = tf.ones([2, 2, 2, 1]);
x = tf.tensor4d([1, 2, 3, 4, 5, 6, 7, 8, 9, 1, 2, 3, 4, 5, 6, 7, 8, 9], inputShape);
dy = tf.tensor4d([3, 1, 2, 0, 3, 1, 2, 0], [2, 2, 2, 1]);
fusedGrads = tf.grads(function (x, w, b) { return tf.fused.depthwiseConv2d({
x: x,
filter: w,
strides: strides,
pad: pad,
dataFormat: 'NHWC',
dilations: [1, 1],
bias: b
}); });
_a = fusedGrads([x, filter, bias], dy), dxFused = _a[0], dfilterFused = _a[1], dbiasFused = _a[2];
grads = tf.grads(function (x, filter, bias) {
var conv = tf.depthwiseConv2d(x, filter, strides, pad);
var sum = tf.add(conv, bias);
return sum;
});
_b = grads([x, filter, bias], dy), dx = _b[0], dfilter = _b[1], dbias = _b[2];
_c = test_util_1.expectArraysClose;
return [4 /*yield*/, dxFused.array()];
case 1:
_d = [_j.sent()];
return [4 /*yield*/, dx.array()];
case 2:
_c.apply(void 0, _d.concat([_j.sent()]));
_e = test_util_1.expectArraysClose;
return [4 /*yield*/, dfilterFused.array()];
case 3:
_f = [_j.sent()];
return [4 /*yield*/, dfilter.array()];
case 4:
_e.apply(void 0, _f.concat([_j.sent()]));
_g = test_util_1.expectArraysClose;
return [4 /*yield*/, dbiasFused.array()];
case 5:
_h = [_j.sent()];
return [4 /*yield*/, dbias.array()];
case 6:
_g.apply(void 0, _h.concat([_j.sent()]));
return [2 /*return*/];
}
});
}); });
it('gradient x=[2,3,3,1] f=[2,2,1,1] s=1 p=0 with bias and activation', function () { return __awaiter(_this, void 0, void 0, function () {
var inputDepth, outputDepth, inputShape, filterSize, strides, pad, filterShape, filter, bias, x, dy, fusedGrads, _a, dxFused, dfilterFused, dbiasFused, grads, _b, dx, dfilter, dbias, _c, _d, _e, _f, _g, _h;
return __generator(this, function (_j) {
switch (_j.label) {
case 0:
inputDepth = 1;
outputDepth = 1;
inputShape = [2, 3, 3, inputDepth];
filterSize = 2;
strides = 1;
pad = 0;
filterShape = [filterSize, filterSize, inputDepth, outputDepth];
filter = tf.tensor4d([-1, 1, -2, 0.5], filterShape);
bias = tf.ones([2, 2, 2, 1]);
x = tf.tensor4d([1, 2, 3, 4, 5, 6, 7, 8, 9, 1, 2, 3, 4, 5, 6, 7, 8, 9], inputShape);
dy = tf.tensor4d([3, 1, 2, 0, 3, 1, 2, 0], [2, 2, 2, 1]);
fusedGrads = tf.grads(function (x, w, b) { return tf.fused.depthwiseConv2d({
x: x,
filter: w,
strides: strides,
pad: pad,
dataFormat: 'NHWC',
dilations: [1, 1],
bias: b,
activation: 'relu'
}); });
_a = fusedGrads([x, filter, bias], dy), dxFused = _a[0], dfilterFused = _a[1], dbiasFused = _a[2];
grads = tf.grads(function (x, filter, bias) {
var conv = tf.depthwiseConv2d(x, filter, strides, pad);
var sum = tf.add(conv, bias);
return tf.relu(sum);
});
_b = grads([x, filter, bias], dy), dx = _b[0], dfilter = _b[1], dbias = _b[2];
_c = test_util_1.expectArraysClose;
return [4 /*yield*/, dxFused.array()];
case 1:
_d = [_j.sent()];
return [4 /*yield*/, dx.array()];
case 2:
_c.apply(void 0, _d.concat([_j.sent()]));
_e = test_util_1.expectArraysClose;
return [4 /*yield*/, dfilterFused.array()];
case 3:
_f = [_j.sent()];
return [4 /*yield*/, dfilter.array()];
case 4:
_e.apply(void 0, _f.concat([_j.sent()]));
_g = test_util_1.expectArraysClose;
return [4 /*yield*/, dbiasFused.array()];
case 5:
_h = [_j.sent()];
return [4 /*yield*/, dbias.array()];
case 6:
_g.apply(void 0, _h.concat([_j.sent()]));
return [2 /*return*/];
}
});
}); });
});
jasmine_util_1.describeWithFlags('fused conv2d', jasmine_util_1.ALL_ENVS, function () {
it('basic', function () { return __awaiter(_this, void 0, void 0, function () {
var inputDepth, inShape, outputDepth, fSize, pad, stride, x, w, result, expected, _a;
return __generator(this, function (_b) {
switch (_b.label) {
case 0:
inputDepth = 2;
inShape = [2, 2, 2, inputDepth];
outputDepth = 2;
fSize = 1;
pad = 0;
stride = 1;
x = tf.tensor4d([1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16], inShape);
w = tf.tensor4d([-1, 1, -2, 0.5], [fSize, fSize, inputDepth, outputDepth]);
result = tf.fused.conv2d({ x: x, filter: w, strides: stride, pad: pad });
expect(result.shape).toEqual([2, 2, 2, 2]);
expected = [-5, 2, -11, 5, -17, 8, -23, 11, -29, 14, -35, 17, -41, 20, -47, 23];
_a = test_util_1.expectArraysClose;
return [4 /*yield*/, result.data()];
case 1:
_a.apply(void 0, [_b.sent(), expected]);
return [2 /*return*/];
}
});
}); });
it('basic with relu', function () { return __awaiter(_this, void 0, void 0, function () {
var inputDepth, inShape, outputDepth, fSize, pad, stride, x, w, result, expected, _a;
return __generator(this, function (_b) {
switch (_b.label) {
case 0:
inputDepth = 2;
inShape = [2, 2, 2, inputDepth];
outputDepth = 2;
fSize = 1;
pad = 0;
stride = 1;
x = tf.tensor4d([1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16], inShape);
w = tf.tensor4d([-1, 1, -2, 0.5], [fSize, fSize, inputDepth, outputDepth]);
result = tf.fused.conv2d({
x: x,
filter: w,
strides: stride,
pad: pad,
dataFormat: 'NHWC',
dilations: [1, 1],
activation: 'relu'
});
expect(result.shape).toEqual([2, 2, 2, 2]);
expected = [0, 2, 0, 5, 0, 8, 0, 11, 0, 14, 0, 17, 0, 20, 0, 23];
_a = test_util_1.expectArraysClose;
return [4 /*yield*/, result.data()];
case 1:
_a.apply(void 0, [_b.sent(), expected]);
return [2 /*return*/];
}
});
}); });
it('basic with bias', function () { return __awaiter(_this, void 0, void 0, function () {
var inputDepth, inShape, outputDepth, fSize, pad, stride, x, w, result, expected, _a;
return __generator(this, function (_b) {
switch (_b.label) {
case 0:
inputDepth = 2;
inShape = [2, 2, 2, inputDepth];
outputDepth = 2;
fSize = 1;
pad = 0;
stride = 1;
x = tf.tensor4d([1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16], inShape);
w = tf.tensor4d([-1, 1, -2, 0.5], [fSize, fSize, inputDepth, outputDepth]);
result = tf.fused.conv2d({
x: x,
filter: w,
strides: stride,
pad: pad,
dataFormat: 'NHWC',
dilations: [1, 1],
bias: tf.tensor1d([5, 6])
});
expect(result.shape).toEqual([2, 2, 2, 2]);
expected = [0, 8, -6, 11, -12, 14, -18, 17, -24, 20, -30, 23, -36, 26, -42, 29];
_a = test_util_1.expectArraysClose;
return [4 /*yield*/, result.data()];
case 1:
_a.apply(void 0, [_b.sent(), expected]);
return [2 /*return*/];
}
});
}); });
it('basic with elu', function () { return __awaiter(_this, void 0, void 0, function () {
var inputDepth, inShape, outputDepth, fSize, pad, stride, x, w, result, expected, _a;
return __generator(this, function (_b) {
switch (_b.label) {
case 0:
inputDepth = 2;
inShape = [2, 2, 2, inputDepth];
outputDepth = 2;
fSize = 1;
pad = 0;
stride = 1;
x = tf.tensor4d([1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16], inShape);
w = tf.tensor4d([-1, 1, -2, 0.5], [fSize, fSize, inputDepth, outputDepth]);
result = tf.fused.conv2d({
x: x,
filter: w,
strides: stride,
pad: pad,
dataFormat: 'NHWC',
dilations: [1, 1],
activation: 'elu'
});
expect(result.shape).toEqual([2, 2, 2, 2]);
expected = [-0.99326, 2, -1, 5, -1, 8, -1, 11, -1, 14, -1, 17, -1, 20, -1, 23];
_a = test_util_1.expectArraysClose;
return [4 /*yield*/, result.data()];
case 1:
_a.apply(void 0, [_b.sent(), expected]);
return [2 /*return*/];
}
});
}); });
it('basic with prelu', function () { return __awaiter(_this, void 0, void 0, function () {
var inputDepth, inShape, outputDepth, fSize, pad, stride, x, alpha, w, result, expected, _a;
return __generator(this, function (_b) {
switch (_b.label) {
case 0:
inputDepth = 2;
inShape = [2, 2, 2, inputDepth];
outputDepth = 2;
fSize = 1;
pad = 0;
stride = 1;
x = tf.tensor4d([1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16], inShape);
alpha = tf.tensor3d([0.25, 0.75], [1, 1, 2]);
w = tf.tensor4d([-1, 1, -2, 0.5], [fSize, fSize, inputDepth, outputDepth]);
result = tf.fused.conv2d({
x: x,
filter: w,
strides: stride,
pad: pad,
dataFormat: 'NHWC',
dilations: [1, 1],
activation: 'prelu',
preluActivationWeights: alpha
});
expect(result.shape).toEqual([2, 2, 2, 2]);
expected = [
-1.25, 2, -2.75, 5, -4.25, 8, -5.75, 11, -7.25, 14, -8.75, 17, -10.25, 20,
-11.75, 23
];
_a = test_util_1.expectArraysClose;
return [4 /*yield*/, result.data()];
case 1:
_a.apply(void 0, [_b.sent(), expected]);
return [2 /*return*/];
}
});
}); });
it('basic with broadcasted bias and relu', function () { return __awaiter(_this, void 0, void 0, function () {
var inputDepth, inShape, outputDepth, fSize, pad, stride, x, w, result, expected, _a;
return __generator(this, function (_b) {
switch (_b.label) {
case 0:
inputDepth = 2;
inShape = [2, 2, 2, inputDepth];
outputDepth = 2;
fSize = 1;
pad = 0;
stride = 1;
x = tf.tensor4d([1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16], inShape);
w = tf.tensor4d([-1, 1, -2, 0.5], [fSize, fSize, inputDepth, outputDepth]);
result = tf.fused.conv2d({
x: x,
filter: w,
strides: stride,
pad: pad,
dataFormat: 'NHWC',
dilations: [1, 1],
bias: tf.scalar(5),
activation: 'relu'
});
expect(result.shape).toEqual([2, 2, 2, 2]);
expected = [0, 7, 0, 10, 0, 13, 0, 16, 0, 19, 0, 22, 0, 25, 0, 28];
_a = test_util_1.expectArraysClose;
return [4 /*yield*/, result.data()];
case 1:
_a.apply(void 0, [_b.sent(), expected]);
return [2 /*return*/];
}
});
}); });
it('im2row', function () { return __awaiter(_this, void 0, void 0, function () {
var inputDepth, inputShape, outputDepth, fSize, pad, strides, x, w, result, _a;
return __generator(this, function (_b) {
switch (_b.label) {
case 0:
inputDepth = 1;
inputShape = [4, 4, inputDepth];
outputDepth = 3;
fSize = 1;
pad = 'same';
strides = [2, 2];
x = tf.tensor3d([
10, 30, 50, 70, 20, 40, 60, 80, -10, -30, -50, -70, -20, -40, -60, -80
], inputShape);
w = tf.tensor4d([1, 0.5, 1], [fSize, fSize, inputDepth, outputDepth]);
result = tf.fused.conv2d({ x: x, filter: w, strides: strides, pad: pad });
_a = test_util_1.expectArraysClose;
return [4 /*yield*/, result.data()];
case 1:
_a.apply(void 0, [_b.sent(),
[10, 5, 10, 50, 25, 50, -10, -5, -10, -50, -25, -50]]);
return [2 /*return*/];
}
});
}); });
it('im2row with relu', function () { return __awaiter(_this, void 0, void 0, function () {
var inputDepth, inputShape, outputDepth, fSize, pad, strides, x, w, result, _a;
return __generator(this, function (_b) {
switch (_b.label) {
case 0:
inputDepth = 1;
inputShape = [4, 4, inputDepth];
outputDepth = 3;
fSize = 1;
pad = 'same';
strides = [2, 2];
x = tf.tensor3d([
10, 30, 50, 70, 20, 40, 60, 80, -10, -30, -50, -70, -20, -40, -60, -80
], inputShape);
w = tf.tensor4d([1, 0.5, 1], [fSize, fSize, inputDepth, outputDepth]);
result = tf.fused.conv2d({
x: x,
filter: w,
strides: strides,
pad: pad,
dataFormat: 'NHWC',
dilations: [1, 1],
activation: 'relu'
});
_a = test_util_1.expectArraysClose;
return [4 /*yield*/, result.data()];
case 1:
_a.apply(void 0, [_b.sent(), [10, 5, 10, 50, 25, 50, 0, 0, 0, 0, 0, 0]]);
return [2 /*return*/];
}
});
}); });
it('im2row with prelu', function () { return __awaiter(_this, void 0, void 0, function () {
var inputDepth, inputShape, ou