@tensorflow/tfjs-core
Version:
Hardware-accelerated JavaScript library for machine intelligence
157 lines • 6.92 kB
JavaScript
/**
* @license
* Copyright 2020 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('mirrorPad', ALL_ENVS, () => {
it('MirrorPad tensor1d', async () => {
const a = tf.tensor1d([1, 2, 3], 'int32');
let b = tf.mirrorPad(a, [[2, 2]], 'reflect');
expectArraysClose(await b.data(), [3, 2, 1, 2, 3, 2, 1]);
expect(b.shape).toEqual([7]);
b = tf.mirrorPad(a, [[2, 2]], 'symmetric');
expectArraysClose(await b.data(), [2, 1, 1, 2, 3, 3, 2]);
expect(b.shape).toEqual([7]);
});
it('MirrorPad tensor2d', async () => {
const a = tf.tensor2d([[1, 2, 3], [4, 5, 6]], [2, 3], 'int32');
let b = tf.mirrorPad(a, [[1, 1], [1, 1]], 'reflect');
// 5, 4, 5, 6, 5
// 2, 1, 2, 3, 2
// 5, 4, 5, 6, 5
// 2, 1, 2, 3, 2
expectArraysClose(await b.data(), [5, 4, 5, 6, 5, 2, 1, 2, 3, 2, 5, 4, 5, 6, 5, 2, 1, 2, 3, 2]);
expect(b.shape).toEqual([4, 5]);
b = tf.mirrorPad(a, [[1, 1], [1, 1]], 'symmetric');
// 1, 1, 2, 3, 3
// 1, 1, 2, 3, 3
// 4, 4, 5, 6, 6
// 4, 4, 5, 6, 6
expectArraysClose(await b.data(), [1, 1, 2, 3, 3, 1, 1, 2, 3, 3, 4, 4, 5, 6, 6, 4, 4, 5, 6, 6]);
expect(b.shape).toEqual([4, 5]);
});
it('MirrorPad tensor3d', async () => {
const a = tf.tensor3d([[[1, 2]], [[3, 4]]], [2, 1, 2], 'int32');
let b = tf.mirrorPad(a, [[1, 1], [0, 0], [1, 1]], 'reflect');
// 4, 3, 4, 3
// 2, 1, 2, 1
// 4, 3, 4, 3
// 2, 1, 2, 1
expectArraysClose(await b.data(), [4, 3, 4, 3, 2, 1, 2, 1, 4, 3, 4, 3, 2, 1, 2, 1]);
expect(b.shape).toEqual([4, 1, 4]);
b = tf.mirrorPad(a, [[1, 1], [0, 0], [1, 1]], 'symmetric');
// 1, 1, 2, 2
// 1, 1, 2, 2
// 3, 3, 4, 4
// 3, 3, 4, 4
expectArraysClose(await b.data(), [1, 1, 2, 2, 1, 1, 2, 2, 3, 3, 4, 4, 3, 3, 4, 4]);
expect(b.shape).toEqual([4, 1, 4]);
});
it('MirrorPad tensor4d', async () => {
const a = tf.tensor4d([[[[1, 2, 3, 4]]]], [1, 1, 1, 4], 'int32');
let b = tf.mirrorPad(a, [[0, 0], [0, 0], [0, 0], [1, 1]], 'reflect');
let expected = tf.tensor4d([[[[2, 1, 2, 3, 4, 3]]]], [1, 1, 1, 6], 'int32');
expectArraysClose(await b.data(), await expected.data());
expect(b.dtype).toBe('int32');
expect(b.shape).toEqual([1, 1, 1, 6]);
b = tf.mirrorPad(a, [[0, 0], [0, 0], [0, 0], [1, 1]], 'symmetric');
expected = tf.tensor4d([[[[1, 1, 2, 3, 4, 4]]]], [1, 1, 1, 6], 'int32');
expectArraysClose(await b.data(), await expected.data());
expect(b.shape).toEqual([1, 1, 1, 6]);
});
it('throws when passed a non-tensor', () => {
expect(() => tf.mirrorPad({}, [[0, 0]], 'reflect'))
.toThrowError(/Argument 'x' passed to 'mirrorPad' must be a Tensor/);
});
it('does not leak memory', () => {
const a = tf.tensor4d([[[[1, 2, 3, 4]]]], [1, 1, 1, 4], 'int32');
// The first call to mirrorPad may create and keeps internal
// singleton tensors. Subsequent calls should always create exactly
// one new tensor.
tf.mirrorPad(a, [[0, 0], [0, 0], [0, 0], [1, 1]], 'reflect');
// Count before real call.
const numTensors = tf.memory().numTensors;
tf.mirrorPad(a, [[0, 0], [0, 0], [0, 0], [1, 1]], 'reflect');
expect(tf.memory().numTensors).toEqual(numTensors + 1);
});
it('accepts a tensor-like object', async () => {
const x = [[1, 2, 3], [4, 5, 6]];
const res = tf.mirrorPad(x, [[1, 1], [1, 1]], 'reflect');
// 5, 4, 5, 6, 5
// 2, 1, 2, 3, 2
// 5, 4, 5, 6, 5
// 2, 1, 2, 3, 2
expectArraysClose(await res.data(), [5, 4, 5, 6, 5, 2, 1, 2, 3, 2, 5, 4, 5, 6, 5, 2, 1, 2, 3, 2]);
expect(res.shape).toEqual([4, 5]);
});
it('Should handle invalid paddings', () => {
const a = tf.tensor1d([1, 2, 3, 4], 'int32');
const f = () => {
// tslint:disable-next-line:no-any
tf.mirrorPad(a, [2, 2, 2], 'reflect');
};
expect(f).toThrowError();
});
it('Should handle paddings that are out of range', () => {
const a = tf.tensor2d([[1, 2, 3], [4, 5, 6]], [2, 3], 'int32');
let f = () => {
// tslint:disable-next-line:no-any
tf.mirrorPad(a, [[4, 1], [1, 1]], 'reflect');
};
expect(f).toThrowError();
f = () => {
// tslint:disable-next-line:no-any
tf.mirrorPad(a, [[-1, 1], [1, 1]], 'reflect');
};
expect(f).toThrowError();
f = () => {
// tslint:disable-next-line:no-any
tf.mirrorPad(a, [[2, 1], [1, 1]], 'reflect');
};
expect(f).toThrowError();
f = () => {
// tslint:disable-next-line:no-any
tf.mirrorPad(a, [[3, 1], [1, 1]], 'symmetric');
};
expect(f).toThrowError();
});
it('Should handle NaNs', async () => {
const a = tf.tensor2d([[1, NaN], [1, NaN]], [2, 2]);
const b = tf.mirrorPad(a, [[1, 1], [1, 1]], 'reflect');
// NaN, 1, NaN, 1
// NaN, 1, NaN, 1
// NaN, 1, NaN, 1
// NaN, 1, NaN, 1
expectArraysClose(await b.data(), [NaN, 1, NaN, 1, NaN, 1, NaN, 1, NaN, 1, NaN, 1, NaN, 1, NaN, 1]);
expect(b.shape).toEqual([4, 4]);
});
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.mirrorPad(a, [[2, 1]], 'reflect'))(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.mirrorPad(a.clone(), [[2, 1]], 'reflect').clone())(a, dy);
expect(da.shape).toEqual([3]);
expectArraysClose(await da.data(), [30, 40, 50]);
});
});
//# sourceMappingURL=mirror_pad_test.js.map