UNPKG

@tensorflow/tfjs-core

Version:

Hardware-accelerated JavaScript library for machine intelligence

325 lines 14.6 kB
/** * @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('where', ALL_ENVS, () => { it('Scalars.', async () => { const a = tf.scalar(10); const b = tf.scalar(20); const c = tf.scalar(1, 'bool'); expectArraysClose(await tf.where(c, a, b).data(), [10]); }); it('Invalid condition type', () => { const c = tf.tensor1d([1, 0, 1, 0], 'int32'); const a = tf.tensor1d([10, 10, 10, 10], 'bool'); const b = tf.tensor1d([20, 20, 20, 20], 'bool'); const f = () => { tf.where(c, a, b); }; expect(f).toThrowError(); }); it('Tensor1D', async () => { const c = tf.tensor1d([1, 0, 1, 0], 'bool'); const a = tf.tensor1d([10, 10, 10, 10]); const b = tf.tensor1d([20, 20, 20, 20]); expectArraysClose(await tf.where(c, a, b).data(), [10, 20, 10, 20]); }); it('Tensor1D different a/b shapes', () => { let c = tf.tensor1d([1, 0, 1, 0], 'bool'); let a = tf.tensor1d([10, 10, 10]); let b = tf.tensor1d([20, 20, 20, 20]); let f = () => { tf.where(c, a, b); }; expect(f).toThrowError(); c = tf.tensor1d([1, 0, 1, 0], 'bool'); a = tf.tensor1d([10, 10, 10, 10]); b = tf.tensor1d([20, 20, 20]); f = () => { tf.where(c, a, b); }; }); it('Tensor1D different condition/a shapes', () => { const c = tf.tensor1d([1, 0, 1, 0], 'bool'); const a = tf.tensor1d([10, 10, 10]); const b = tf.tensor1d([20, 20, 20]); const f = () => { tf.where(c, a, b); }; expect(f).toThrowError(); }); it('Tensor2D', async () => { const c = tf.tensor2d([[1, 0], [0, 1]], [2, 2], 'bool'); const a = tf.tensor2d([[10, 10], [10, 10]], [2, 2]); const b = tf.tensor2d([[5, 5], [5, 5]], [2, 2]); expectArraysClose(await tf.where(c, a, b).data(), [10, 5, 5, 10]); }); it('Tensor2D different a/b shapes', () => { let c = tf.tensor2d([[1, 1], [0, 0]], [2, 2], 'bool'); let a = tf.tensor2d([[5, 5, 5], [5, 5, 5]], [2, 3]); let b = tf.tensor2d([[4, 4], [4, 4]], [2, 2]); let f = () => { tf.where(c, a, b); }; expect(f).toThrowError(); c = tf.tensor2d([[1, 1], [0, 0]], [2, 2], 'bool'); a = tf.tensor2d([[5, 5], [5, 5]], [2, 2]); b = tf.tensor2d([[4, 4, 4], [4, 4, 4]], [2, 3]); f = () => { tf.where(c, a, b); }; expect(f).toThrowError(); }); it('Tensor2D different condition/a shapes', () => { const c = tf.tensor2d([[1, 0], [0, 1]], [2, 2], 'bool'); const a = tf.tensor2d([[10, 10, 10], [10, 10, 10]], [2, 3]); const b = tf.tensor2d([[5, 5, 5], [5, 5, 5]], [2, 3]); const f = () => { tf.where(c, a, b); }; expect(f).toThrowError(); }); it('Tensor2D different `a` dimension w/ condition rank=1', async () => { const c = tf.tensor1d([1, 0, 1, 0], 'bool'); let a = tf.tensor2d([[10, 10], [10, 10]], [2, 2]); let b = tf.tensor2d([[5, 5], [5, 5]], [2, 2]); const f = () => { tf.where(c, a, b); }; expect(f).toThrowError(); a = tf.tensor2d([[10], [10], [10], [10]], [4, 1]); b = tf.tensor2d([[5], [5], [5], [5]], [4, 1]); expectArraysClose(await tf.where(c, a, b).data(), [10, 5, 10, 5]); a = tf.tensor2d([[10, 10], [10, 10], [10, 10], [10, 10]], [4, 2]); b = tf.tensor2d([[5, 5], [5, 5], [5, 5], [5, 5]], [4, 2]); expectArraysClose(await tf.where(c, a, b).data(), [10, 10, 5, 5, 10, 10, 5, 5]); }); it('Tensor3D', async () => { const c = tf.tensor3d([[[1], [0], [1]], [[0], [0], [0]]], [2, 3, 1], 'bool'); const a = tf.tensor3d([[[5], [5], [5]], [[5], [5], [5]]], [2, 3, 1]); const b = tf.tensor3d([[[3], [3], [3]], [[3], [3], [3]]], [2, 3, 1]); expectArraysClose(await tf.where(c, a, b).data(), [5, 3, 5, 3, 3, 3]); }); it('Tensor3D with scalar condition', async () => { const a = tf.ones([1, 3, 3]); const b = tf.zeros([1, 3, 3]); expectArraysClose(await tf.where(tf.ones([1], 'bool'), a, b).data(), [1, 1, 1, 1, 1, 1, 1, 1, 1]); expectArraysClose(await tf.where(tf.zeros([1], 'bool'), a, b).data(), [0, 0, 0, 0, 0, 0, 0, 0, 0]); }); it('1D condition with higher rank a and b', async () => { const condition = tf.tensor1d([1, 0, 0, 1, 1], 'bool'); const a = tf.ones([5, 2, 2]); const b = tf.fill([5, 2, 2], 3); expectArraysClose(await tf.where(condition, a, b).data(), [1, 1, 1, 1, 3, 3, 3, 3, 3, 3, 3, 3, 1, 1, 1, 1, 1, 1, 1, 1]); }); it('Tensor3D different a/b shapes', () => { const c = tf.tensor3d([[[1], [0], [1]], [[0], [0], [0]]], [2, 3, 1], 'bool'); let a = tf.tensor3d([[[5], [5]], [[5], [5]]], [2, 2, 1]); let b = tf.tensor3d([[[3], [3], [3]], [[3], [3], [3]]], [2, 3, 1]); let f = () => { tf.where(c, a, b); }; expect(f).toThrowError(); a = tf.tensor3d([[[5], [5], [5]], [[5], [5], [5]]], [2, 3, 1]); b = tf.tensor3d([[[3], [3]], [[3], [3]]], [2, 2, 1]); f = () => { tf.where(c, a, b); }; expect(f).toThrowError(); }); it('Tensor3D different condition/a shapes', () => { const c = tf.tensor3d([[[1], [0]], [[0], [0]]], [2, 2, 1], 'bool'); const a = tf.tensor3d([[[5], [5], [5]], [[5], [5], [5]]], [2, 3, 1]); const b = tf.tensor3d([[[3], [3], [3]], [[3], [3], [3]]], [2, 3, 1]); const f = () => { tf.where(c, a, b); }; expect(f).toThrowError(); }); it('Tensor3D different `a` dimension w/ condition rank=1', async () => { const c = tf.tensor1d([1, 0, 1, 0], 'bool'); let a = tf.tensor3d([[[9, 9], [9, 9]], [[9, 9], [9, 9]]], [2, 2, 2]); let b = tf.tensor3d([[[8, 8], [8, 8]], [[8, 8], [8, 8]]], [2, 2, 2]); const f = () => { tf.where(c, a, b); }; expect(f).toThrowError(); a = tf.tensor3d([[[9]], [[9]], [[9]], [[9]]], [4, 1, 1]); b = tf.tensor3d([[[8]], [[8]], [[8]], [[8]]], [4, 1, 1]); expectArraysClose(await tf.where(c, a, b).data(), [9, 8, 9, 8]); a = tf.tensor3d([[[9], [9]], [[9], [9]], [[9], [9]], [[9], [9]]], [4, 2, 1]); b = tf.tensor3d([[[8], [8]], [[8], [8]], [[8], [8]], [[8], [8]]], [4, 2, 1]); expectArraysClose(await tf.where(c, a, b).data(), [9, 9, 8, 8, 9, 9, 8, 8]); }); it('Tensor4D', async () => { const c = tf.tensor4d([1, 0, 1, 1], [2, 2, 1, 1], 'bool'); const a = tf.tensor4d([7, 7, 7, 7], [2, 2, 1, 1]); const b = tf.tensor4d([3, 3, 3, 3], [2, 2, 1, 1]); expectArraysClose(await tf.where(c, a, b).data(), [7, 3, 7, 7]); }); it('Tensor4D different a/b shapes', () => { const c = tf.tensor4d([1, 0, 1, 1], [2, 2, 1, 1], 'bool'); let a = tf.tensor4d([7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7], [2, 3, 2, 1]); let b = tf.tensor4d([3, 3, 3, 3], [2, 2, 1, 1]); let f = () => { tf.where(c, a, b); }; expect(f).toThrowError(); a = tf.tensor4d([7, 7, 7, 7], [2, 2, 1, 1]); b = tf.tensor4d([3, 3, 3, 3, 3, 3, 3, 3], [2, 2, 2, 1]); f = () => { tf.where(c, a, b); }; expect(f).toThrowError(); }); it('Tensor4D broadcastable a/b shapes', () => { const c = tf.tensor4d([1, 0, 1, 1], [2, 2, 1, 1], 'bool'); const a = tf.tensor4d([7, 7, 7, 7, 7, 7, 7, 7], [2, 2, 2, 1]); const b = [3]; let f = () => { tf.where(c, a, b); }; expect(f).toThrowError(); const a1 = [7]; const b1 = tf.tensor4d([3, 3, 3, 3, 3, 3, 3, 3], [2, 2, 2, 1]); f = () => { tf.where(c, a1, b1); }; expect(f).toThrowError(); }); it('Tensor4D different condition/a shapes', () => { const c = tf.tensor4d([1, 0, 1, 1, 1, 0, 1, 1], [2, 2, 2, 1], 'bool'); const a = tf.tensor4d([7, 7, 7, 7], [2, 2, 1, 1]); const b = tf.tensor4d([3, 3, 3, 3], [2, 2, 1, 1]); const f = () => { tf.where(c, a, b); }; expect(f).toThrowError(); }); it('Tensor4D different `a` dimension w/ condition rank=1', async () => { const c = tf.tensor1d([1, 0, 1, 0], 'bool'); let a = tf.tensor4d([7, 7, 7, 7, 7, 7, 7, 7], [2, 2, 2, 1]); let b = tf.tensor4d([3, 3, 3, 3, 3, 3, 3, 3], [2, 2, 2, 1]); const f = () => { tf.where(c, a, b); }; expect(f).toThrowError(); a = tf.tensor4d([7, 7, 7, 7], [4, 1, 1, 1]); b = tf.tensor4d([3, 3, 3, 3], [4, 1, 1, 1]); expectArraysClose(await tf.where(c, a, b).data(), [7, 3, 7, 3]); a = tf.tensor4d([7, 7, 7, 7, 7, 7, 7, 7], [4, 2, 1, 1]); b = tf.tensor4d([3, 3, 3, 3, 3, 3, 3, 3], [4, 2, 1, 1]); expectArraysClose(await tf.where(c, a, b).data(), [7, 7, 3, 3, 7, 7, 3, 3]); }); it('TensorLike', async () => { expectArraysClose(await tf.where(true, 10, 20).data(), [10]); }); it('TensorLike Chained', async () => { const a = tf.scalar(10); expectArraysClose(await a.where(true, 20).data(), [10]); }); it('throws when passed condition as a non-tensor', () => { expect(() => tf.where({}, tf.scalar(1, 'bool'), tf.scalar(1, 'bool'))) .toThrowError(/Argument 'condition' passed to 'where' must be a Tensor/); }); it('throws when passed a as a non-tensor', () => { expect(() => tf.where(tf.scalar(1, 'bool'), {}, tf.scalar(1, 'bool'))) .toThrowError(/Argument 'a' passed to 'where' must be a Tensor/); }); it('throws when passed b as a non-tensor', () => { expect(() => tf.where(tf.scalar(1, 'bool'), tf.scalar(1, 'bool'), {})) .toThrowError(/Argument 'b' passed to 'where' must be a Tensor/); }); it('accepts a tensor-like object', async () => { const a = 10; const b = 20; const c = 1; expectArraysClose(await tf.where(c, a, b).data(), [10]); }); it('1D gradient', async () => { const c = tf.tensor1d([1, 0, 1], 'bool'); const a = tf.tensor1d([1, 2, 3]); const b = tf.tensor1d([4, 5, 6]); const dy = tf.tensor1d([1, 2, 3]); const grads = tf.grads((c, a, b) => tf.where(c, a, b)); const [dc, da, db] = grads([c, a, b], dy); expectArraysClose(await dc.data(), [0, 0, 0]); expectArraysClose(await da.data(), [1, 0, 3]); expectArraysClose(await db.data(), [0, 2, 0]); expect(dc.shape).toEqual(c.shape); expect(da.shape).toEqual(a.shape); expect(db.shape).toEqual(b.shape); }); it('gradient with clones', async () => { const c = tf.tensor1d([1, 0, 1], 'bool'); const a = tf.tensor1d([1, 2, 3]); const b = tf.tensor1d([4, 5, 6]); const dy = tf.tensor1d([1, 2, 3]); const grads = tf.grads((c, a, b) => tf.where(c.clone(), a.clone(), b.clone()).clone()); const [dc, da, db] = grads([c, a, b], dy); expectArraysClose(await dc.data(), [0, 0, 0]); expectArraysClose(await da.data(), [1, 0, 3]); expectArraysClose(await db.data(), [0, 2, 0]); expect(dc.shape).toEqual(c.shape); expect(da.shape).toEqual(a.shape); expect(db.shape).toEqual(b.shape); }); it('2D gradient', async () => { const c = tf.tensor2d([1, 0, 1, 1, 1, 0], [2, 3], 'bool'); const a = tf.tensor2d([1, 2, 3, 4, 5, 6], [2, 3]); const b = tf.tensor2d([7, 8, 9, 10, 11, 12], [2, 3]); const dy = tf.tensor2d([1, 2, 3, 4, 5, 6], [2, 3]); const grads = tf.grads((c, a, b) => tf.where(c, a, b)); const [dc, da, db] = grads([c, a, b], dy); expectArraysClose(await dc.data(), [0, 0, 0, 0, 0, 0]); expectArraysClose(await da.data(), [1, 0, 3, 4, 5, 0]); expectArraysClose(await db.data(), [0, 2, 0, 0, 0, 6]); expect(dc.shape).toEqual(c.shape); expect(da.shape).toEqual(a.shape); expect(db.shape).toEqual(b.shape); }); it('3D gradient', async () => { const c = tf.tensor3d([1, 1, 0, 1, 1, 0], [2, 3, 1], 'bool'); const a = tf.tensor3d([1, 2, 3, 4, 5, 6], [2, 3, 1]); const b = tf.tensor3d([7, 8, 9, 10, 11, 12], [2, 3, 1]); const dy = tf.tensor3d([1, 2, 3, 4, 5, 6], [2, 3, 1]); const grads = tf.grads((c, a, b) => tf.where(c, a, b)); const [dc, da, db] = grads([c, a, b], dy); expectArraysClose(await dc.data(), [0, 0, 0, 0, 0, 0]); expectArraysClose(await da.data(), [1, 2, 0, 4, 5, 0]); expectArraysClose(await db.data(), [0, 0, 3, 0, 0, 6]); expect(dc.shape).toEqual(c.shape); expect(da.shape).toEqual(a.shape); expect(db.shape).toEqual(b.shape); }); it('4D gradient', async () => { const c = tf.tensor4d([1, 1, 0, 1], [2, 2, 1, 1], 'bool'); const a = tf.tensor4d([1, 2, 3, 4], [2, 2, 1, 1]); const b = tf.tensor4d([5, 6, 7, 8], [2, 2, 1, 1]); const dy = tf.tensor4d([1, 2, 3, 4], [2, 2, 1, 1]); const grads = tf.grads((c, a, b) => tf.where(c, a, b)); const [dc, da, db] = grads([c, a, b], dy); expectArraysClose(await dc.data(), [0, 0, 0, 0]); expectArraysClose(await da.data(), [1, 2, 0, 4]); expectArraysClose(await db.data(), [0, 0, 3, 0]); expect(dc.shape).toEqual(c.shape); expect(da.shape).toEqual(a.shape); expect(db.shape).toEqual(b.shape); }); }); //# sourceMappingURL=where_test.js.map