UNPKG

@tensorflow/tfjs-core

Version:

Hardware-accelerated JavaScript library for machine intelligence

368 lines (339 loc) 12.3 kB
/** * @license * Copyright 2018 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. * ============================================================================= */ /** * Linear algebra ops. */ import {ENGINE} from '../engine'; import {dispose} from '../globals'; import {Tensor, Tensor1D, Tensor2D} from '../tensor'; import {convertToTensor} from '../tensor_util_env'; import {TensorLike} from '../types'; import {assert} from '../util'; import {squeeze, stack, unstack} from './array_ops'; import {sub} from './binary_ops'; import {split} from './concat_split'; import {eye} from './eye'; import {logicalAnd, where} from './logical_ops'; import {norm} from './norm'; import {op} from './operation'; import {sum} from './reduction_ops'; import {range, scalar, tensor2d, zeros} from './tensor_ops'; /** * Copy a tensor setting everything outside a central band in each innermost * matrix to zero. * * The band part is computed as follows: Assume input has `k` dimensions * `[I, J, K, ..., M, N]`, then the output is a tensor with the same shape where * `band[i, j, k, ..., m, n] = in_band(m, n) * input[i, j, k, ..., m, n]`. * The indicator function * `in_band(m, n) = (num_lower < 0 || (m-n) <= num_lower))` * `&& (num_upper < 0 || (n-m) <= num_upper)` * * ```js * const x = tf.tensor2d([[ 0, 1, 2, 3], * [-1, 0, 1, 2], * [-2, -1, 0, 1], * [-3, -2, -1, 0]]); * let y = tf.linalg.bandPart(x, 1, -1); * y.print(); // [[ 0, 1, 2, 3], * // [-1, 0, 1, 2], * // [ 0, -1, 0, 1], * // [ 0, 0 , -1, 0]] * let z = tf.linalg.bandPart(x, 2, 1); * z.print(); // [[ 0, 1, 0, 0], * // [-1, 0, 1, 0], * // [-2, -1, 0, 1], * // [ 0, -2, -1, 0]] * ``` * * @param x Rank `k` tensor * @param numLower Number of subdiagonals to keep. * If negative, keep entire lower triangle. * @param numUpper Number of subdiagonals to keep. * If negative, keep entire upper triangle. * @returns Rank `k` tensor of the same shape as input. * The extracted banded tensor. */ /** * @doc {heading:'Operations', * subheading:'Linear Algebra', * namespace:'linalg'} */ function bandPart_<T extends Tensor>( a: T|TensorLike, numLower: number, numUpper: number): T { if (numLower % 1 !== 0) { throw new Error( `bandPart(): numLower must be an integer, got ${numLower}.`); } if (numUpper % 1 !== 0) { throw new Error( `bandPart(): numUpper must be an integer, got ${numUpper}.`); } const $a = convertToTensor(a, 'a', 'bandPart'); if ($a.rank < 2) { throw new Error(`bandPart(): Rank must be at least 2, got ${$a.rank}.`); } const shape = $a.shape, [M, N] = $a.shape.slice(-2); if (!(numLower <= M)) { throw new Error( `bandPart(): numLower (${numLower})` + ` must not be greater than the number of rows (${M}).`); } if (!(numUpper <= N)) { throw new Error( `bandPart(): numUpper (${numUpper})` + ` must not be greater than the number of columns (${N}).`); } if (numLower < 0) { numLower = M; } if (numUpper < 0) { numUpper = N; } const i = range(0, M, 1, 'int32').reshape([-1, 1]), j = range(0, N, 1, 'int32'), ij = sub(i, j); const inBand = logicalAnd( ij.lessEqual(scalar(+numLower, 'int32')), ij.greaterEqual(scalar(-numUpper, 'int32'))); const zero = zeros([M, N], $a.dtype); return stack(unstack($a.reshape([-1, M, N])) .map(mat => where(inBand, mat, zero))) .reshape(shape) as T; } /** * Gram-Schmidt orthogonalization. * * ```js * const x = tf.tensor2d([[1, 2], [3, 4]]); * let y = tf.linalg.gramSchmidt(x); * y.print(); * console.log('Othogonalized:'); * y.dot(y.transpose()).print(); // should be nearly the identity matrix. * console.log('First row direction maintained:'); * const data = await y.array(); * console.log(data[0][1] / data[0][0]); // should be nearly 2. * ``` * * @param xs The vectors to be orthogonalized, in one of the two following * formats: * - An Array of `tf.Tensor1D`. * - A `tf.Tensor2D`, i.e., a matrix, in which case the vectors are the rows * of `xs`. * In each case, all the vectors must have the same length and the length * must be greater than or equal to the number of vectors. * @returns The orthogonalized and normalized vectors or matrix. * Orthogonalization means that the vectors or the rows of the matrix * are orthogonal (zero inner products). Normalization means that each * vector or each row of the matrix has an L2 norm that equals `1`. */ /** * @doc {heading:'Operations', * subheading:'Linear Algebra', * namespace:'linalg'} */ function gramSchmidt_(xs: Tensor1D[]|Tensor2D): Tensor1D[]|Tensor2D { let inputIsTensor2D: boolean; if (Array.isArray(xs)) { inputIsTensor2D = false; assert( xs != null && xs.length > 0, () => 'Gram-Schmidt process: input must not be null, undefined, or ' + 'empty'); const dim = xs[0].shape[0]; for (let i = 1; i < xs.length; ++i) { assert( xs[i].shape[0] === dim, () => 'Gram-Schmidt: Non-unique lengths found in the input vectors: ' + `(${(xs as Tensor1D[])[i].shape[0]} vs. ${dim})`); } } else { inputIsTensor2D = true; xs = split(xs, xs.shape[0], 0).map(x => squeeze(x, [0])); } assert( xs.length <= xs[0].shape[0], () => `Gram-Schmidt: Number of vectors (${ (xs as Tensor1D[]).length}) exceeds ` + `number of dimensions (${(xs as Tensor1D[])[0].shape[0]}).`); const ys: Tensor1D[] = []; const xs1d = xs; for (let i = 0; i < xs.length; ++i) { ys.push(ENGINE.tidy(() => { let x = xs1d[i]; if (i > 0) { for (let j = 0; j < i; ++j) { const proj = sum(ys[j].mulStrict(x)).mul(ys[j]); x = x.sub(proj); } } return x.div(norm(x, 'euclidean')); })); } if (inputIsTensor2D) { return stack(ys, 0) as Tensor2D; } else { return ys; } } /** * Compute QR decomposition of m-by-n matrix using Householder transformation. * * Implementation based on * [http://www.cs.cornell.edu/~bindel/class/cs6210-f09/lec18.pdf] * (http://www.cs.cornell.edu/~bindel/class/cs6210-f09/lec18.pdf) * * ```js * const a = tf.tensor2d([[1, 2], [3, 4]]); * let [q, r] = tf.linalg.qr(a); * console.log('Q'); * q.print(); * console.log('R'); * r.print(); * console.log('Orthogonalized'); * q.dot(q.transpose()).print() // should be nearly the identity matrix. * console.log('Reconstructed'); * q.dot(r).print(); // should be nearly [[1, 2], [3, 4]]; * ``` * * @param x The `tf.Tensor` to be QR-decomposed. Must have rank >= 2. Suppose * it has the shape `[..., M, N]`. * @param fullMatrices An optional boolean parameter. Defaults to `false`. * If `true`, compute full-sized `Q`. If `false` (the default), * compute only the leading N columns of `Q` and `R`. * @returns An `Array` of two `tf.Tensor`s: `[Q, R]`. `Q` is a unitary matrix, * i.e., its columns all have unit norm and are mutually orthogonal. * If `M >= N`, * If `fullMatrices` is `false` (default), * - `Q` has a shape of `[..., M, N]`, * - `R` has a shape of `[..., N, N]`. * If `fullMatrices` is `true` (default), * - `Q` has a shape of `[..., M, M]`, * - `R` has a shape of `[..., M, N]`. * If `M < N`, * - `Q` has a shape of `[..., M, M]`, * - `R` has a shape of `[..., M, N]`. * @throws If the rank of `x` is less than 2. */ /** * @doc {heading:'Operations', * subheading:'Linear Algebra', * namespace:'linalg'} */ function qr_(x: Tensor, fullMatrices = false): [Tensor, Tensor] { if (x.rank < 2) { throw new Error( `qr() requires input tensor to have a rank >= 2, but got rank ${ x.rank}`); } else if (x.rank === 2) { return qr2d(x as Tensor2D, fullMatrices); } else { // Rank > 2. // TODO(cais): Below we split the input into individual 2D tensors, // perform QR decomposition on them and then stack the results back // together. We should explore whether this can be parallelized. const outerDimsProd = x.shape.slice(0, x.shape.length - 2) .reduce((value, prev) => value * prev); const x2ds = unstack( x.reshape([ outerDimsProd, x.shape[x.shape.length - 2], x.shape[x.shape.length - 1] ]), 0); const q2ds: Tensor2D[] = []; const r2ds: Tensor2D[] = []; x2ds.forEach(x2d => { const [q2d, r2d] = qr2d(x2d as Tensor2D, fullMatrices); q2ds.push(q2d); r2ds.push(r2d); }); const q = stack(q2ds, 0).reshape(x.shape); const r = stack(r2ds, 0).reshape(x.shape); return [q, r]; } } function qr2d(x: Tensor2D, fullMatrices = false): [Tensor2D, Tensor2D] { return ENGINE.tidy(() => { if (x.shape.length !== 2) { throw new Error( `qr2d() requires a 2D Tensor, but got a ${x.shape.length}D Tensor.`); } const m = x.shape[0]; const n = x.shape[1]; let q = eye(m); // Orthogonal transform so far. let r = x.clone(); // Transformed matrix so far. const one2D = tensor2d([[1]], [1, 1]); let w: Tensor2D = one2D.clone(); const iters = m >= n ? n : m; for (let j = 0; j < iters; ++j) { // This tidy within the for-loop ensures we clean up temporary // tensors as soon as they are no longer needed. const rTemp = r; const wTemp = w; const qTemp = q; [w, r, q] = ENGINE.tidy((): [Tensor2D, Tensor2D, Tensor2D] => { // Find H = I - tau * w * w', to put zeros below R(j, j). const rjEnd1 = r.slice([j, j], [m - j, 1]); const normX = rjEnd1.norm(); const rjj = r.slice([j, j], [1, 1]); // The sign() function returns 0 on 0, which causes division by zero. const s = tensor2d([[-1]]).where(rjj.greater(0), tensor2d([[1]])); const u1 = rjj.sub(s.mul(normX)); const wPre = rjEnd1.div(u1); if (wPre.shape[0] === 1) { w = one2D.clone(); } else { w = one2D.concat( wPre.slice([1, 0], [wPre.shape[0] - 1, wPre.shape[1]]) as Tensor2D, 0); } const tau = s.matMul(u1).div(normX).neg() as Tensor2D; // -- R := HR, Q := QH. const rjEndAll = r.slice([j, 0], [m - j, n]); const tauTimesW: Tensor2D = tau.mul(w); const wT: Tensor2D = w.transpose(); if (j === 0) { r = rjEndAll.sub(tauTimesW.matMul(wT.matMul(rjEndAll))); } else { const rTimesTau: Tensor2D = rjEndAll.sub(tauTimesW.matMul(wT.matMul(rjEndAll))); r = r.slice([0, 0], [j, n]).concat(rTimesTau, 0); } const tawTimesWT: Tensor2D = tauTimesW.transpose(); const qAllJEnd = q.slice([0, j], [m, q.shape[1] - j]); if (j === 0) { q = qAllJEnd.sub(qAllJEnd.matMul(w).matMul(tawTimesWT)); } else { const qTimesTau: Tensor2D = qAllJEnd.sub(qAllJEnd.matMul(w).matMul(tawTimesWT)); q = q.slice([0, 0], [m, j]).concat(qTimesTau, 1); } return [w, r, q]; }); dispose([rTemp, wTemp, qTemp]); } if (!fullMatrices && m > n) { q = q.slice([0, 0], [m, n]); r = r.slice([0, 0], [n, n]); } return [q, r]; }) as [Tensor2D, Tensor2D]; } export const bandPart = op({bandPart_}); export const gramSchmidt = op({gramSchmidt_}); export const qr = op({qr_});