"use strict";
|
/**
|
* @license
|
* Copyright 2017 Google Inc. 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 ops_1 = require("./ops/ops");
|
var tape_1 = require("./tape");
|
var test_util_1 = require("./test_util");
|
jasmine_util_1.describeWithFlags('getFilteredNodesXToY', jasmine_util_1.ALL_ENVS, function () {
|
it('no paths from x to y', function () {
|
var x = tf.scalar(1);
|
var intermediate1 = tf.scalar(0);
|
var intermediate2 = tf.scalar(0);
|
var y = tf.scalar(2);
|
var tape = [
|
{
|
id: 0,
|
kernelName: 'node0',
|
inputs: { x: x },
|
outputs: [intermediate1],
|
gradient: null
|
},
|
{
|
id: 1,
|
kernelName: 'node1',
|
inputs: { intermediate2: intermediate2 },
|
outputs: [y],
|
gradient: null
|
}
|
];
|
var filteredTapeNodes = tape_1.getFilteredNodesXToY(tape, [x], y);
|
expect(filteredTapeNodes.length).toBe(0);
|
expect(filteredTapeNodes).toEqual([]);
|
});
|
it('one operation x => y', function () {
|
var x = tf.scalar(1);
|
var y = tf.scalar(2);
|
var tape = [
|
{ id: 0, kernelName: 'node0', inputs: { x: x }, outputs: [y], gradient: null }
|
];
|
var filteredTapeNodes = tape_1.getFilteredNodesXToY(tape, [x], y);
|
expect(filteredTapeNodes.length).toBe(1);
|
expect(filteredTapeNodes).toEqual(tape);
|
});
|
it('1 operation [x0, x1] => y, all input paths', function () {
|
var x0 = tf.scalar(0);
|
var x1 = tf.scalar(1);
|
var y = tf.scalar(2);
|
var tape = [{
|
id: 0,
|
kernelName: 'node0',
|
inputs: { x0: x0, x1: x1 },
|
outputs: [y],
|
gradient: null
|
}];
|
var filteredTapeNodes = tape_1.getFilteredNodesXToY(tape, [x0, x1], y);
|
expect(filteredTapeNodes.length).toBe(1);
|
expect(filteredTapeNodes).toEqual(tape);
|
});
|
it('one operation [x0, x1] => y, one input paths', function () {
|
var x0 = tf.scalar(0);
|
var x1 = tf.scalar(1);
|
var y = tf.scalar(2);
|
var tape = [{
|
id: 0,
|
kernelName: 'node0',
|
inputs: { x0: x0, x1: x1 },
|
outputs: [y],
|
gradient: null
|
}];
|
var filteredTapeNodes = tape_1.getFilteredNodesXToY(tape, [x0], y);
|
expect(filteredTapeNodes.length).toBe(1);
|
// x1 input should be pruned, we don't ask for the gradient of x1.
|
expect(filteredTapeNodes[0]).toEqual({
|
id: 0,
|
kernelName: 'node0',
|
inputs: { x0: x0 },
|
outputs: [y],
|
gradient: null
|
});
|
});
|
it('two operations x => intermediate => y', function () {
|
var x = tf.scalar(1);
|
var intermediate = tf.scalar(0);
|
var y = tf.scalar(2);
|
var tape = [
|
{
|
id: 0,
|
kernelName: 'node0',
|
inputs: { x: x },
|
outputs: [intermediate],
|
gradient: null
|
},
|
{
|
id: 1,
|
kernelName: 'node1',
|
inputs: { intermediate: intermediate },
|
outputs: [y],
|
gradient: null
|
}
|
];
|
var filteredTapeNodes = tape_1.getFilteredNodesXToY(tape, [x], y);
|
expect(filteredTapeNodes.length).toBe(2);
|
expect(filteredTapeNodes).toEqual(tape);
|
});
|
it('two operations [x0, x1], [x2] => ' +
|
'intermediate => y', function () {
|
var x0 = tf.scalar(1);
|
var x1 = tf.scalar(2);
|
var x2 = tf.scalar(3);
|
var intermediate = tf.scalar(4);
|
var y = tf.scalar(2);
|
var tape = [
|
{
|
id: 0,
|
kernelName: 'node0',
|
inputs: { x0: x0, x1: x1 },
|
outputs: [intermediate],
|
gradient: null
|
},
|
{
|
id: 1,
|
kernelName: 'node1',
|
inputs: { x2: x2, intermediate: intermediate },
|
outputs: [y],
|
gradient: null
|
}
|
];
|
var filteredTapeNodes = tape_1.getFilteredNodesXToY(tape, [x0, x1, x2], y);
|
expect(filteredTapeNodes.length).toBe(2);
|
expect(filteredTapeNodes).toEqual(tape);
|
});
|
it('x => y and x => orphan', function () {
|
var x = tf.scalar(1);
|
var orphan = tf.scalar(0);
|
var y = tf.scalar(2);
|
var tape = [
|
{
|
id: 0,
|
kernelName: 'node0',
|
inputs: { x: x },
|
outputs: [orphan],
|
gradient: null
|
},
|
{ id: 1, kernelName: 'node1', inputs: { x: x }, outputs: [y], gradient: null }
|
];
|
var filteredTapeNodes = tape_1.getFilteredNodesXToY(tape, [x], y);
|
expect(filteredTapeNodes.length).toBe(1);
|
// The orphan should be removed.
|
expect(filteredTapeNodes[0]).toEqual(tape[1]);
|
});
|
it('x => y and orphan => y', function () {
|
var x = tf.scalar(1);
|
var orphan = tf.scalar(0);
|
var y = tf.scalar(2);
|
var tape = [{
|
id: 0,
|
kernelName: 'node0',
|
inputs: { x: x, orphan: orphan },
|
outputs: [y],
|
gradient: null
|
}];
|
var filteredTapeNodes = tape_1.getFilteredNodesXToY(tape, [x], y);
|
expect(filteredTapeNodes.length).toBe(1);
|
// The orphan should be pruned from the node's input.
|
expect(filteredTapeNodes[0]).toEqual({
|
id: 0,
|
kernelName: 'node0',
|
inputs: { x: x },
|
outputs: [y],
|
gradient: null
|
});
|
});
|
it('1 op with 3 outputs x => y1, y2, y3', function () {
|
var x = tf.scalar(1);
|
var y1 = tf.scalar(2);
|
var y2 = tf.scalar(2);
|
var y3 = tf.scalar(2);
|
var tape = [{
|
id: 0,
|
kernelName: 'node0',
|
inputs: { x: x },
|
outputs: [y1, y2, y3],
|
gradient: null
|
}];
|
var filteredNodes1 = tape_1.getFilteredNodesXToY(tape, [x], y1);
|
expect(filteredNodes1.length).toBe(1);
|
expect(filteredNodes1).toEqual(tape);
|
var filteredNodes2 = tape_1.getFilteredNodesXToY(tape, [x], y2);
|
expect(filteredNodes2.length).toBe(1);
|
expect(filteredNodes2).toEqual(tape);
|
var filteredNodes3 = tape_1.getFilteredNodesXToY(tape, [x], y3);
|
expect(filteredNodes3.length).toBe(1);
|
expect(filteredNodes3).toEqual(tape);
|
});
|
});
|
jasmine_util_1.describeWithFlags('backpropagateGradients', jasmine_util_1.ALL_ENVS, function () {
|
it('Throws if gradient is not defined', function () {
|
var x = tf.scalar(0);
|
var y = tf.scalar(1);
|
var dy = tf.scalar(1);
|
var accumulatedGradientsMap = {};
|
accumulatedGradientsMap[y.id] = dy;
|
var tape = [
|
{ id: 0, kernelName: 'node0', inputs: { x: x }, outputs: [y], gradient: null }
|
];
|
expect(function () { return tape_1.backpropagateGradients(accumulatedGradientsMap, tape, function (f) { return tf.tidy(f); }); })
|
.toThrowError();
|
});
|
it('basic backprop with 1 node', function () { return __awaiter(_this, void 0, void 0, function () {
|
var x, y, dy, accumulatedGradientsMap, tape, _a;
|
return __generator(this, function (_b) {
|
switch (_b.label) {
|
case 0:
|
x = tf.scalar(0);
|
y = tf.scalar(1);
|
dy = tf.scalar(1);
|
accumulatedGradientsMap = {};
|
accumulatedGradientsMap[y.id] = dy;
|
tape = [{
|
id: 0,
|
kernelName: 'node0',
|
inputs: { x: x },
|
outputs: [y],
|
gradient: function (dys) {
|
return { x: function () { return dys[0].add(tf.scalar(1)); } };
|
}
|
}];
|
tape_1.backpropagateGradients(accumulatedGradientsMap, tape, function (f) { return tf.tidy(f); });
|
_a = test_util_1.expectArraysClose;
|
return [4 /*yield*/, accumulatedGradientsMap[x.id].data()];
|
case 1:
|
_a.apply(void 0, [_b.sent(), [2]]);
|
return [2 /*return*/];
|
}
|
});
|
}); });
|
it('basic backprop with 2 nodes', function () { return __awaiter(_this, void 0, void 0, function () {
|
var x, intermediate, y, dy, accumulatedGradientsMap, tape, _a;
|
return __generator(this, function (_b) {
|
switch (_b.label) {
|
case 0:
|
x = tf.scalar(0);
|
intermediate = tf.scalar(1);
|
y = tf.scalar(2);
|
dy = tf.scalar(1);
|
accumulatedGradientsMap = {};
|
accumulatedGradientsMap[y.id] = dy;
|
tape = [
|
{
|
id: 0,
|
kernelName: 'node0',
|
inputs: { x: x },
|
outputs: [intermediate],
|
gradient: function (dys) {
|
return { x: function () { return dys[0].add(tf.scalar(1)); } };
|
}
|
},
|
{
|
id: 1,
|
kernelName: 'node1',
|
inputs: { intermediate: intermediate },
|
outputs: [y],
|
gradient: function (dys) {
|
return { intermediate: function () { return dys[0].add(tf.scalar(1)); } };
|
}
|
}
|
];
|
tape_1.backpropagateGradients(accumulatedGradientsMap, tape, function (f) { return tf.tidy(f); });
|
// dx = dy + 1 + 1
|
_a = test_util_1.expectArraysClose;
|
return [4 /*yield*/, accumulatedGradientsMap[x.id].data()];
|
case 1:
|
// dx = dy + 1 + 1
|
_a.apply(void 0, [_b.sent(), [3]]);
|
return [2 /*return*/];
|
}
|
});
|
}); });
|
it('basic backprop with a split node accumulates gradients', function () { return __awaiter(_this, void 0, void 0, function () {
|
var x, intermediate1, intermediate2, y, dy, accumulatedGradientsMap, tape, _a, _b;
|
return __generator(this, function (_c) {
|
switch (_c.label) {
|
case 0:
|
x = tf.scalar(0);
|
intermediate1 = tf.scalar(1);
|
intermediate2 = tf.scalar(2);
|
y = tf.scalar(3);
|
dy = tf.scalar(1);
|
accumulatedGradientsMap = {};
|
accumulatedGradientsMap[y.id] = dy;
|
tape = [
|
{
|
id: 0,
|
kernelName: 'node0',
|
inputs: { x: x },
|
outputs: [intermediate1],
|
gradient: function (dys) {
|
return { x: function () { return dys[0].add(tf.scalar(1)); } };
|
}
|
},
|
{
|
id: 1,
|
kernelName: 'node1',
|
inputs: { x: x },
|
outputs: [intermediate2],
|
gradient: function (dys) {
|
return { x: function () { return dys[0].add(tf.scalar(1)); } };
|
}
|
},
|
{
|
id: 2,
|
kernelName: 'node2',
|
inputs: { intermediate1: intermediate1, intermediate2: intermediate2 },
|
outputs: [y],
|
gradient: function (dys) {
|
return {
|
intermediate1: function () { return dys[0].add(tf.scalar(1)); },
|
intermediate2: function () { return dys[0].add(tf.scalar(1)); }
|
};
|
}
|
}
|
];
|
tape_1.backpropagateGradients(accumulatedGradientsMap, tape, function (f) { return tf.tidy(f); });
|
// dx = dy + 1 + 1 + 1 + 1 + 1
|
_a = test_util_1.expectArraysClose;
|
return [4 /*yield*/, accumulatedGradientsMap[x.id].data()];
|
case 1:
|
_b = [_c.sent()];
|
return [4 /*yield*/, dy.data()];
|
case 2:
|
// dx = dy + 1 + 1 + 1 + 1 + 1
|
_a.apply(void 0, _b.concat([[(_c.sent())[0] + 5]]));
|
return [2 /*return*/];
|
}
|
});
|
}); });
|
it('backprop over 1 node with 3 outputs, w.r.t to the 2nd output', function () { return __awaiter(_this, void 0, void 0, function () {
|
var x, y1, y2, y3, accumulatedGradientsMap, dy2, dys, tape, _a, _b, _c, _d;
|
return __generator(this, function (_e) {
|
switch (_e.label) {
|
case 0:
|
x = tf.tensor1d([1, 1, 1]);
|
y1 = tf.scalar(1);
|
y2 = tf.scalar(1);
|
y3 = tf.scalar(1);
|
accumulatedGradientsMap = {};
|
dy2 = tf.scalar(5);
|
accumulatedGradientsMap[y2.id] = dy2;
|
tape = [{
|
id: 0,
|
kernelName: 'node0',
|
inputs: { x: x },
|
outputs: [y1, y2, y3],
|
gradient: function (dys_) {
|
dys = dys_.map(function (dy) { return dy || ops_1.zerosLike(y1); });
|
return { x: function () { return tf.stack(dys); } };
|
}
|
}];
|
tape_1.backpropagateGradients(accumulatedGradientsMap, tape, function (f) { return tf.tidy(f); });
|
_a = test_util_1.expectArraysClose;
|
return [4 /*yield*/, accumulatedGradientsMap[x.id].data()];
|
case 1:
|
_a.apply(void 0, [_e.sent(), [0, 5, 0]]);
|
_b = test_util_1.expectArraysClose;
|
return [4 /*yield*/, dys[0].data()];
|
case 2:
|
_b.apply(void 0, [_e.sent(), [0]]);
|
_c = test_util_1.expectArraysClose;
|
return [4 /*yield*/, dys[1].data()];
|
case 3:
|
_c.apply(void 0, [_e.sent(), [5]]);
|
_d = test_util_1.expectArraysClose;
|
return [4 /*yield*/, dys[2].data()];
|
case 4:
|
_d.apply(void 0, [_e.sent(), [0]]);
|
return [2 /*return*/];
|
}
|
});
|
}); });
|
});
|
//# sourceMappingURL=tape_test.js.map
|
