| 1 | /**
|
| 2 | * @license
|
| 3 | * Copyright 2020 Google LLC. All Rights Reserved.
|
| 4 | * Licensed under the Apache License, Version 2.0 (the "License");
|
| 5 | * you may not use this file except in compliance with the License.
|
| 6 | * You may obtain a copy of the License at
|
| 7 | *
|
| 8 | * http://www.apache.org/licenses/LICENSE-2.0
|
| 9 | *
|
| 10 | * Unless required by applicable law or agreed to in writing, software
|
| 11 | * distributed under the License is distributed on an "AS IS" BASIS,
|
| 12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
| 13 | * See the License for the specific language governing permissions and
|
| 14 | * limitations under the License.
|
| 15 | * =============================================================================
|
| 16 | */
|
| 17 | import { ENGINE } from '../engine';
|
| 18 | import { Tile } from '../kernel_names';
|
| 19 | import { convertToTensor } from '../tensor_util_env';
|
| 20 | import { clone } from './clone';
|
| 21 | import { op } from './operation';
|
| 22 | import { reshape } from './reshape';
|
| 23 | /**
|
| 24 | * Broadcast an array to a compatible shape NumPy-style.
|
| 25 | *
|
| 26 | * The tensor's shape is compared to the broadcast shape from end to beginning.
|
| 27 | * Ones are prepended to the tensor's shape until is has the same length as
|
| 28 | * the broadcast shape. If input.shape[i]==shape[i], the (i+1)-th axis is
|
| 29 | * already broadcast-compatible. If input.shape[i]==1 and shape[i]==N, then
|
| 30 | * the input tensor is tiled N times along that axis (using tf.tile).
|
| 31 | *
|
| 32 | * @param input The tensor that is to be broadcasted.
|
| 33 | * @param shape The input is to be broadcast to this shape.
|
| 34 | *
|
| 35 | * @doc {heading: 'Tensors', subheading: 'Transformations'}
|
| 36 | */
|
| 37 | function broadcastTo_(x, shape) {
|
| 38 | let input = convertToTensor(x, 'broadcastTo', 'x');
|
| 39 | const xShape = input.shape;
|
| 40 | if (shape.some(d => !(d > 0) || d % 1 !== 0)) {
|
| 41 | throw new Error(`broadcastTo(): Invalid broadcast shape [${shape}].`);
|
| 42 | }
|
| 43 | if (shape.length < input.rank) {
|
| 44 | throw new Error(`broadcastTo(): shape.length=${shape.length} < input.rank=${input.rank}.`);
|
| 45 | }
|
| 46 | if (shape.length > input.rank) {
|
| 47 | const newShape = input.shape.slice();
|
| 48 | while (newShape.length < shape.length) {
|
| 49 | newShape.unshift(1);
|
| 50 | }
|
| 51 | input = reshape(input, newShape);
|
| 52 | }
|
| 53 | const inputShape = input.shape;
|
| 54 | const reps = Array.from(shape);
|
| 55 | for (let i = shape.length - 1; i >= 0; i--) {
|
| 56 | if (inputShape[i] === shape[i]) {
|
| 57 | reps[i] = 1;
|
| 58 | }
|
| 59 | else if (input.shape[i] !== 1) {
|
| 60 | throw new Error(`broadcastTo(): [${xShape}] cannot be broadcast to [${shape}].`);
|
| 61 | }
|
| 62 | }
|
| 63 | const axes = reps.map((n, i) => n > 1 ? i : -1).filter(i => i >= 0);
|
| 64 | if (axes.length === 0) {
|
| 65 | return clone(input);
|
| 66 | }
|
| 67 | // TODO call broadcastTo kernel directly once backends implement broadcstTo
|
| 68 | const inputs = { x: input };
|
| 69 | const attrs = { reps };
|
| 70 | return ENGINE.runKernel(Tile, inputs, attrs);
|
| 71 | }
|
| 72 | export const broadcastTo = op({ broadcastTo_ });
|
| 73 | //# sourceMappingURL=broadcast_to.js.map |
| \ | No newline at end of file |