@tensorflow/tfjs-core
Version:
Hardware-accelerated JavaScript library for machine intelligence
245 lines • 10.2 kB
JavaScript
/**
* @license
* Copyright 2017 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.
* =============================================================================
*/
import * as tf from '../index';
import { ALL_ENVS, describeWithFlags } from '../jasmine_util';
import { expectArraysClose } from '../test_util';
describeWithFlags('pad 1d', ALL_ENVS, () => {
it('Should pad 1D arrays', async () => {
const a = tf.tensor1d([1, 2, 3, 4, 5, 6], 'int32');
const b = tf.pad1d(a, [2, 3]);
expectArraysClose(await b.data(), [0, 0, 1, 2, 3, 4, 5, 6, 0, 0, 0]);
});
it('Should not pad 1D arrays with 0s', async () => {
const a = tf.tensor1d([1, 2, 3, 4], 'int32');
const b = tf.pad1d(a, [0, 0]);
expectArraysClose(await b.data(), [1, 2, 3, 4]);
});
it('Should handle padding with custom value', async () => {
let a = tf.tensor1d([1, 2, 3, 4], 'int32');
let b = tf.pad1d(a, [2, 3], 9);
expectArraysClose(await b.data(), [9, 9, 1, 2, 3, 4, 9, 9, 9]);
a = tf.tensor1d([1, 2, 3, 4]);
b = tf.pad1d(a, [2, 1], 1.1);
expectArraysClose(await b.data(), [1.1, 1.1, 1, 2, 3, 4, 1.1]);
a = tf.tensor1d([1, 2, 3, 4]);
b = tf.pad1d(a, [2, 1], 1);
expectArraysClose(await b.data(), [1, 1, 1, 2, 3, 4, 1]);
a = tf.tensor1d([1, 2, 3, 4]);
b = tf.pad1d(a, [2, 1], Number.NEGATIVE_INFINITY);
expectArraysClose(await b.data(), [
Number.NEGATIVE_INFINITY, Number.NEGATIVE_INFINITY, 1, 2, 3, 4,
Number.NEGATIVE_INFINITY
]);
a = tf.tensor1d([1, 2, 3, 4]);
b = tf.pad1d(a, [2, 1], Number.POSITIVE_INFINITY);
expectArraysClose(await b.data(), [
Number.POSITIVE_INFINITY, Number.POSITIVE_INFINITY, 1, 2, 3, 4,
Number.POSITIVE_INFINITY
]);
});
it('Should handle NaNs with 1D arrays', async () => {
const a = tf.tensor1d([1, NaN, 2, NaN]);
const b = tf.pad1d(a, [1, 1]);
expectArraysClose(await b.data(), [0, 1, NaN, 2, NaN, 0]);
});
it('Should handle invalid paddings', () => {
const a = tf.tensor1d([1, 2, 3, 4], 'int32');
const f = () => {
// tslint:disable-next-line:no-any
tf.pad1d(a, [2, 2, 2]);
};
expect(f).toThrowError();
});
it('grad', async () => {
const a = tf.tensor1d([1, 2, 3]);
const dy = tf.tensor1d([10, 20, 30, 40, 50, 60]);
const da = tf.grad((a) => tf.pad1d(a, [2, 1]))(a, dy);
expect(da.shape).toEqual([3]);
expectArraysClose(await da.data(), [30, 40, 50]);
});
it('gradient with clones', async () => {
const a = tf.tensor1d([1, 2, 3]);
const dy = tf.tensor1d([10, 20, 30, 40, 50, 60]);
const da = tf.grad((a) => tf.pad1d(a.clone(), [2, 1]).clone())(a, dy);
expect(da.shape).toEqual([3]);
expectArraysClose(await da.data(), [30, 40, 50]);
});
it('accepts a tensor-like object', async () => {
const a = [1, 2, 3, 4, 5, 6];
const b = tf.pad1d(a, [2, 3]);
expectArraysClose(await b.data(), [0, 0, 1, 2, 3, 4, 5, 6, 0, 0, 0]);
});
});
describeWithFlags('pad 2d', ALL_ENVS, () => {
it('Should pad 2D arrays', async () => {
let a = tf.tensor2d([[1], [2]], [2, 1], 'int32');
let b = tf.pad2d(a, [[1, 1], [1, 1]]);
// 0, 0, 0
// 0, 1, 0
// 0, 2, 0
// 0, 0, 0
expectArraysClose(await b.data(), [0, 0, 0, 0, 1, 0, 0, 2, 0, 0, 0, 0]);
a = tf.tensor2d([[1, 2, 3], [4, 5, 6]], [2, 3], 'int32');
b = tf.pad2d(a, [[2, 2], [1, 1]]);
// 0, 0, 0, 0, 0
// 0, 0, 0, 0, 0
// 0, 1, 2, 3, 0
// 0, 4, 5, 6, 0
// 0, 0, 0, 0, 0
// 0, 0, 0, 0, 0
expectArraysClose(await b.data(), [
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 2, 3, 0,
0, 4, 5, 6, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
]);
});
it('Should not pad 2D arrays with 0s', async () => {
const a = tf.tensor2d([[1, 2, 3], [4, 5, 6]], [2, 3], 'int32');
const b = tf.pad2d(a, [[0, 0], [0, 0]]);
expectArraysClose(await b.data(), [1, 2, 3, 4, 5, 6]);
});
it('Should handle padding with custom value', async () => {
let a = tf.tensor2d([[1, 2, 3], [4, 5, 6]], [2, 3], 'int32');
let b = tf.pad2d(a, [[1, 1], [1, 1]], 10);
expectArraysClose(await b.data(), [
10, 10, 10, 10, 10, 10, 1, 2, 3, 10,
10, 4, 5, 6, 10, 10, 10, 10, 10, 10
]);
a = tf.tensor2d([[1], [1]], [2, 1]);
b = tf.pad2d(a, [[1, 1], [1, 1]], -2.1);
expectArraysClose(await b.data(), [-2.1, -2.1, -2.1, -2.1, 1, -2.1, -2.1, 1, -2.1, -2.1, -2.1, -2.1]);
a = tf.tensor2d([[1], [1]], [2, 1]);
b = tf.pad2d(a, [[1, 1], [1, 1]], -2);
expectArraysClose(await b.data(), [-2, -2, -2, -2, 1, -2, -2, 1, -2, -2, -2, -2]);
});
it('Should handle NaNs with 2D arrays', async () => {
const a = tf.tensor2d([[1, NaN], [1, NaN]], [2, 2]);
const b = tf.pad2d(a, [[1, 1], [1, 1]]);
// 0, 0, 0, 0
// 0, 1, NaN, 0
// 0, 1, NaN, 0
// 0, 0, 0, 0
expectArraysClose(await b.data(), [0, 0, 0, 0, 0, 1, NaN, 0, 0, 1, NaN, 0, 0, 0, 0, 0]);
});
it('Should handle invalid paddings', () => {
const a = tf.tensor2d([[1], [2]], [2, 1], 'int32');
const f = () => {
// tslint:disable-next-line:no-any
tf.pad2d(a, [[2, 2, 2], [1, 1, 1]]);
};
expect(f).toThrowError();
});
it('grad', async () => {
const a = tf.tensor2d([[1, 2], [3, 4]]);
const dy = tf.tensor2d([[0, 0, 0], [10, 20, 0], [30, 40, 0]], [3, 3]);
const da = tf.grad((a) => tf.pad2d(a, [[1, 0], [0, 1]]))(a, dy);
expect(da.shape).toEqual([2, 2]);
expectArraysClose(await da.data(), [10, 20, 30, 40]);
});
it('accepts a tensor-like object', async () => {
const a = [[1, 2, 3], [4, 5, 6]]; // 2x3
const b = tf.pad2d(a, [[0, 0], [0, 0]]);
expectArraysClose(await b.data(), [1, 2, 3, 4, 5, 6]);
});
});
describeWithFlags('pad 3d', ALL_ENVS, () => {
it('works with 3d tensor, float32', async () => {
const a = tf.tensor3d([[[1]], [[2]]], [2, 1, 1], 'float32');
const b = tf.pad3d(a, [[1, 1], [1, 1], [1, 1]]);
// 0, 0, 0
// 0, 0, 0
// 0, 0, 0
// 0, 0, 0
// 0, 1, 0
// 0, 0, 0
// 0, 0, 0
// 0, 2, 0
// 0, 0, 0
// 0, 0, 0
// 0, 0, 0
// 0, 0, 0
expect(b.shape).toEqual([4, 3, 3]);
expectArraysClose(await b.data(), [
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0,
0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
]);
});
});
describeWithFlags('pad 4d', ALL_ENVS, () => {
it('Should pad 4D arrays', async () => {
const a = tf.tensor4d([[[[9]]]], [1, 1, 1, 1], 'int32');
const b = tf.pad4d(a, [[0, 0], [1, 1], [1, 1], [0, 0]]);
const expected = tf.tensor4d([[[[0], [0], [0]], [[0], [9], [0]], [[0], [0], [0]]]], [1, 3, 3, 1], 'int32');
expectArraysClose(await b.data(), await expected.data());
expect(b.dtype).toBe('int32');
expect(b.shape).toEqual([1, 3, 3, 1]);
});
it('does not leak memory', () => {
const a = tf.tensor4d([[[[9]]]], [1, 1, 1, 1], 'int32');
// The first call to pad may create and keeps internal singleton tensors.
// Subsequent calls should always create exactly one new tensor.
tf.pad4d(a, [[0, 0], [1, 1], [1, 1], [0, 0]]);
// Count before real call.
const numTensors = tf.memory().numTensors;
tf.pad4d(a, [[0, 0], [1, 1], [1, 1], [0, 0]]);
expect(tf.memory().numTensors).toEqual(numTensors + 1);
});
it('accepts a tensor-like object', async () => {
const a = [[[[9]]]]; // 1x1x1x1
const b = tf.pad4d(a, [[0, 0], [1, 1], [1, 1], [0, 0]]);
const expected = tf.tensor4d([[[[0], [0], [0]], [[0], [9], [0]], [[0], [0], [0]]]], [1, 3, 3, 1], 'float32');
expectArraysClose(await b.data(), await expected.data());
expect(b.dtype).toBe('float32');
expect(b.shape).toEqual([1, 3, 3, 1]);
});
});
describeWithFlags('pad', ALL_ENVS, () => {
it('Pad tensor2d', async () => {
let a = tf.tensor2d([[1], [2]], [2, 1], 'int32');
let b = tf.pad(a, [[1, 1], [1, 1]]);
// 0, 0, 0
// 0, 1, 0
// 0, 2, 0
// 0, 0, 0
expectArraysClose(await b.data(), [0, 0, 0, 0, 1, 0, 0, 2, 0, 0, 0, 0]);
a = tf.tensor2d([[1, 2, 3], [4, 5, 6]], [2, 3], 'int32');
b = tf.pad(a, [[2, 2], [1, 1]]);
// 0, 0, 0, 0, 0
// 0, 0, 0, 0, 0
// 0, 1, 2, 3, 0
// 0, 4, 5, 6, 0
// 0, 0, 0, 0, 0
// 0, 0, 0, 0, 0
expectArraysClose(await b.data(), [
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 2, 3, 0,
0, 4, 5, 6, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
]);
});
it('throws when passed a non-tensor', () => {
expect(() => tf.pad({}, [[0, 0]]))
.toThrowError(/Argument 'x' passed to 'pad' must be a Tensor/);
});
it('accepts a tensor-like object', async () => {
const x = [[1], [2]];
const res = tf.pad(x, [[1, 1], [1, 1]]);
// 0, 0, 0
// 0, 1, 0
// 0, 2, 0
// 0, 0, 0
expectArraysClose(await res.data(), [0, 0, 0, 0, 1, 0, 0, 2, 0, 0, 0, 0]);
});
});
//# sourceMappingURL=pad_test.js.map