2
0
Fork 0

add: cve_simple pure C implementation.

This commit is contained in:
Daniel Kapla 2019-09-25 13:53:45 +02:00
parent 099079330a
commit 8db5266f8e
19 changed files with 824 additions and 335 deletions

View File

@ -55,21 +55,21 @@
#' @import stats
#' @importFrom stats model.frame
#' @export
cve <- function(formula, data, method = "simple", max.dim = 10, ...) {
cve <- function(formula, data, method = "simple", max.dim = 10L, ...) {
# check for type of `data` if supplied and set default
if (missing(data)) {
data <- environment(formula)
} else if (!is.data.frame(data)) {
stop('Parameter `data` must be a `data.frame` or missing.')
stop("Parameter 'data' must be a 'data.frame' or missing.")
}
# extract `X`, `Y` from `formula` with `data`
model <- stats::model.frame(formula, data)
X <- as.matrix(model[,-1, drop = FALSE])
Y <- as.matrix(model[, 1, drop = FALSE])
X <- as.matrix(model[ ,-1L, drop = FALSE])
Y <- as.double(model[ , 1L])
# pass extracted data on to [cve.call()]
dr <- cve.call(X, Y, method = method, ...)
dr <- cve.call(X, Y, method = method, max.dim = max.dim, ...)
# overwrite `call` property from [cve.call()]
dr$call <- match.call()
@ -84,35 +84,81 @@ cve <- function(formula, data, method = "simple", max.dim = 10, ...) {
#' @param ... Method specific parameters.
#' @rdname cve
#' @export
cve.call <- function(X, Y, method = "simple", nObs = nrow(X)^.5,
min.dim = 1, max.dim = 10, k, ...) {
cve.call <- function(X, Y, method = "simple",
nObs = sqrt(nrow(X)), h = NULL,
min.dim = 1L, max.dim = 10L, k = NULL,
tau = 1.0, tol = 1e-3,
epochs = 50L, attempts = 10L,
logger = NULL) {
# parameter checking
if (!is.matrix(X)) {
stop('X should be a matrices.')
if (!(is.matrix(X) && is.numeric(X))) {
stop("Parameter 'X' should be a numeric matrices.")
}
if (is.matrix(Y)) {
Y <- as.vector(Y)
if (!is.numeric(Y)) {
stop("Parameter 'Y' must be numeric.")
}
if (is.matrix(Y) || !is.double(Y)) {
Y <- as.double(Y)
}
if (nrow(X) != length(Y)) {
stop('Rows of X and number of Y elements are not compatible.')
stop("Rows of 'X' and 'Y' elements are not compatible.")
}
if (ncol(X) < 2) {
stop('X is one dimensional, no need for dimension reduction.')
stop("'X' is one dimensional, no need for dimension reduction.")
}
if (!missing(k)) {
min.dim <- as.integer(k)
max.dim <- as.integer(k)
} else {
if (missing(k) || is.null(k)) {
min.dim <- as.integer(min.dim)
max.dim <- as.integer(min(max.dim, ncol(X) - 1L))
} else {
min.dim <- as.integer(k)
max.dim <- as.integer(k)
}
if (min.dim > max.dim) {
stop('`min.dim` bigger `max.dim`.')
stop("'min.dim' bigger 'max.dim'.")
}
if (max.dim >= ncol(X)) {
stop('`max.dim` must be smaller than `ncol(X)`.')
stop("'max.dim' (or 'k') must be smaller than 'ncol(X)'.")
}
if (is.function(h)) {
estimate.bandwidth <- h
h <- NULL
}
if (!is.numeric(tau) || length(tau) > 1L || tau <= 0.0) {
stop("Initial step-width 'tau' must be positive number.")
} else {
tau <- as.double(tau)
}
if (!is.numeric(tol) || length(tol) > 1L || tol < 0.0) {
stop("Break condition tolerance 'tol' must be not negative number.")
} else {
tol <- as.double(tol)
}
if (!is.numeric(epochs) || length(epochs) > 1L) {
stop("Parameter 'epochs' must be positive integer.")
} else if (!is.integer(epochs)) {
epochs <- as.integer(epochs)
}
if (epochs < 1L) {
stop("Parameter 'epochs' must be at least 1L.")
}
if (!is.numeric(attempts) || length(attempts) > 1L) {
stop("Parameter 'attempts' must be positive integer.")
} else if (!is.integer(attempts)) {
attempts <- as.integer(attempts)
}
if (attempts < 1L) {
stop("Parameter 'attempts' must be at least 1L.")
}
if (is.function(logger)) {
loggerEnv <- environment(logger)
} else {
loggerEnv <- NULL
}
# Call specified method.
@ -120,15 +166,40 @@ cve.call <- function(X, Y, method = "simple", nObs = nrow(X)^.5,
call <- match.call()
dr <- list()
for (k in min.dim:max.dim) {
if (missing(h) || is.null(h)) {
h <- estimate.bandwidth(X, k, nObs)
} else if (is.numeric(h) && h > 0.0) {
h <- as.double(h)
} else {
stop("Bandwidth 'h' must be positive numeric.")
}
if (method == 'simple') {
dr.k <- cve_simple(X, Y, k, nObs = nObs, ...)
} else if (method == 'linesearch') {
dr.k <- cve_linesearch(X, Y, k, nObs = nObs, ...)
} else if (method == 'sgd') {
dr.k <- cve_sgd(X, Y, k, nObs = nObs, ...)
dr.k <- .Call('cve_simple', PACKAGE = 'CVE',
X, Y, k, h,
tau, tol,
epochs, attempts,
logger, loggerEnv)
# dr.k <- cve_simple(X, Y, k, nObs = nObs, ...)
# } else if (method == 'linesearch') {
# dr.k <- cve_linesearch(X, Y, k, nObs = nObs, ...)
# } else if (method == 'rcg') {
# dr.k <- cve_rcg(X, Y, k, nObs = nObs, ...)
# } else if (method == 'momentum') {
# dr.k <- cve_momentum(X, Y, k, nObs = nObs, ...)
# } else if (method == 'rmsprob') {
# dr.k <- cve_rmsprob(X, Y, k, nObs = nObs, ...)
# } else if (method == 'sgdrmsprob') {
# dr.k <- cve_sgdrmsprob(X, Y, k, nObs = nObs, ...)
# } else if (method == 'sgd') {
# dr.k <- cve_sgd(X, Y, k, nObs = nObs, ...)
} else {
stop('Got unknown method.')
}
dr.k$B <- null(dr.k$V)
dr.k$loss <- mean(dr.k$L)
dr.k$h <- h
dr.k$k <- k
class(dr.k) <- "cve.k"
dr[[k]] <- dr.k
@ -165,43 +236,20 @@ cve.call <- function(X, Y, method = "simple", nObs = nrow(X)^.5,
#' @method plot cve
#' @export
plot.cve <- function(x, ...) {
L <- c()
k <- c()
for (dr.k in x) {
if (class(dr.k) == 'cve.k') {
k <- c(k, paste0(dr.k$k))
L <- c(L, dr.k$L)
}
}
L <- matrix(L, ncol = length(k))
boxplot(L, main = "Loss ...",
xlab = "SDR dimension k",
ylab = expression(L(V, X[i])),
names = k)
# H <- x$history
# H_1 <- H[!is.na(H[, 1]), 1]
# defaults <- list(
# main = "History",
# xlab = "Iterations i",
# ylab = expression(loss == L[n](V^{(i)})),
# xlim = c(1, nrow(H)),
# ylim = c(0, max(H)),
# type = "l"
# )
# call.plot <- match.call()
# keys <- names(defaults)
# keys <- keys[match(keys, names(call.plot)[-1], nomatch = 0) == 0]
# for (key in keys) {
# call.plot[[key]] <- defaults[[key]]
# }
# call.plot[[1L]] <- quote(plot)
# call.plot$x <- quote(1:length(H_1))
# call.plot$y <- quote(H_1)
# eval(call.plot)
# if (ncol(H) > 1) {
# for (i in 2:ncol(H)) {
# H_i <- H[H[, i] > 0, i]
# lines(1:length(H_i), H_i)
# }
# }
# x.ends <- apply(H, 2, function(h) { length(h[!is.na(h)]) })
# y.ends <- apply(H, 2, function(h) { tail(h[!is.na(h)], n=1) })
# points(x.ends, y.ends)
}
#' Prints a summary of a \code{cve} result.

45
CVE_C/src/callLogger.c Normal file
View File

@ -0,0 +1,45 @@
#include "cve.h"
/**
* Calles a R function passed to the algoritm and supplied intermidiate
* optimization values for logging the optimization progress.
* The supplied parameters to the logger functions are as follows:
* - attempt: Attempts counter.
* - epoch: Current epoch staring with 0 as initial epoch.
* - L: Per X_i to X_j pair loss.
* - V: Current estimated SDR null space basis.
* - tau: Step-size.
* - err: Error \eqn{|| V V^T - V_{tau} V_{tau}^T ||}.
*
* @param logger Pointer to a SEXP R object representing an R function.
* @param loggerEnv Pointer to a SEXP R object representing an R environment.
*/
void callLogger(SEXP logger, SEXP env,
const int attempt, const int epoch,
const double* L, const int lenL,
const double* V, const int nrowV, const int ncolV,
const double tau) {
/* Create R objects to be passed to R logger function. */
// Attempt is converted from 0-indexed to 1-indexed as R index.
SEXP R_attempt = PROTECT(ScalarInteger(attempt + 1));
// No index shift for the epoch because the 0 epoch is before the first
// optimization step.
SEXP R_epoch = PROTECT(ScalarInteger(epoch));
SEXP R_L = PROTECT(allocVector(REALSXP, lenL));
SEXP R_V = PROTECT(allocMatrix(REALSXP, nrowV, ncolV));
SEXP R_tau = PROTECT(ScalarReal(tau));
/* Copy data to created R objects. */
memcpy(REAL(R_L), L, lenL * sizeof(double));
memcpy(REAL(R_V), V, nrowV * ncolV * sizeof(double));
/* Create logger function call as R language expression. */
SEXP R_exp = PROTECT(lang6(logger, R_epoch, R_attempt,
R_L, R_V, R_tau));
/* Evaluate the logger function call expression. */
eval(R_exp, env);
/* Unprotext created R objects. */
UNPROTECT(6);
}

View File

@ -1,7 +0,0 @@
#ifndef CVE_INCLUDE_GUARD_CONFIG_
#define CVE_INCLUDE_GUARD_CONFIG_
#define CVE_MEM_CHUNK_SIZE 2032
#define CVE_MEM_CHUNK_SMALL 1016
#endif /* CVE_INCLUDE_GUARD_CONFIG_ */

98
CVE_C/src/cve.h Normal file
View File

@ -0,0 +1,98 @@
/* Include Guard */
#ifndef CVE_INCLUDE_GUARD_H_
#define CVE_INCLUDE_GUARD_H_
#include <string.h> // `mem*` functions.
#include <math.h> // sqrt, exp, ...
#include <R.h>
#include <Rinternals.h>
#include <R_ext/BLAS.h>
#include <R_ext/Lapack.h>
#define CVE_MEM_CHUNK_SIZE 2032
#define CVE_MEM_CHUNK_SMALL 1016
void cve_simple_sub(const int n, const int p, const int q,
const double *X, const double *Y, const double h,
const double tau_init, const double tol_init,
const int epochs, const int attempts,
double *V, double *L,
SEXP logger, SEXP loggerEnv);
void callLogger(SEXP logger, SEXP env,
const int attempt, const int epoch,
const double* L, const int lenL,
const double* V, const int nrowV, const int ncolV,
const double tau);
/* CVE sub-routines */
int getWorkLen(const int n, const int p, const int q);
double cost(const int n,
const double *Y,
const double *vecK,
const double *colSums,
double *y1, double *L);
void scaling(const int n,
const double *Y, const double *y1, const double *L,
const double *vecD, const double *vecK,
const double *colSums,
double *vecS);
/* rStiefl */
void rStiefl(const int p, const int q, double *V,
double *workMem, int workLen);
/* MATRIX */
double norm(const double *A, const int nrow, const int ncol,
const char *type);
void matrixprod(const double *A, const int nrowA, const int ncolA,
const double *B, const int nrowB, const int ncolB,
double *C);
void crossprod(const double *A, const int nrowA, const int ncolA,
const double *B, const int nrowB, const int ncolB,
double *C);
void skew(const int nrow, const int ncol,
double alpha, const double *A, const double *B,
double beta,
double *C);
void nullProj(const double *B, const int nrowB, const int ncolB,
double *Q);
void scale(const double s, double *A, const int nelem);
void cayleyTransform(const int p, const int q,
const double *A, const double *B,
double *X, double *workMem);
/* Row and column opperations. */
void rowSums(const double *A, const int nrow, const int ncol,
double *sum);
void colSums(const double *A, const int nrow, const int ncol,
double *sum);
void rowSquareSums(const double *A, const int nrow, const int ncol,
double *sum);
void rowSumsSymVec(const double *Avec, const int nrow,
const double diag,
double *sum);
void rowDiffs(const double* X, const int nrow, const int ncol,
double *diffs);
void rowDiffSquareSums(const double* X, const int nrow, const int ncol,
double *sum);
/* SWEEP */
void rowSweep(const double *A, const int nrow, const int ncol,
const char* op,
const double *v, // vector of length nrow
double *C);
#endif /* CVE_INCLUDE_GUARD_H_ */

165
CVE_C/src/cve_simple.c Normal file
View File

@ -0,0 +1,165 @@
#include "cve.h"
// TODO: clarify
static inline double gaussKernel(const double x, const double scale) {
return exp(scale * x * x);
}
void cve_simple_sub(const int n, const int p, const int q,
const double *X, const double *Y, const double h,
const double tau_init, const double tol_init,
const int epochs, const int attempts,
double *V, double *L,
SEXP logger, SEXP loggerEnv) {
int attempt, epoch, i, j, k, nn = (n * (n - 1)) / 2;
double loss, loss_last, loss_best, err, tau;
double tol = tol_init * sqrt((double)(2 * q));
double gKscale = -0.5 / h;
/* Create further intermediate or internal variables. */
double *Q = (double*)R_alloc(p * p, sizeof(double));
double *V_best = (double*)R_alloc(p * q, sizeof(double));
double *L_best = (double*)R_alloc(n, sizeof(double));
double *V_tau = (double*)R_alloc(p * q, sizeof(double));
double *X_diff = (double*)R_alloc(nn * p, sizeof(double));
double *X_proj = (double*)R_alloc(nn * p, sizeof(double)); // TODO: needed?
double *y1 = (double*)R_alloc(n , sizeof(double)); // TODO: needed?
double *vecD = (double*)R_alloc(nn, sizeof(double));
double *vecK = (double*)R_alloc(nn, sizeof(double));
double *vecS = (double*)R_alloc(nn, sizeof(double));
double *colSums = (double*)R_alloc(n, sizeof(double));
double *G = (double*)R_alloc(p * q, sizeof(double));
double *A = (double*)R_alloc(p * p, sizeof(double));
/* Determine size of working memory used by subroutines. */
const int workLen = getWorkLen(n, p, q);
double *workMem = (double*)R_alloc(workLen, sizeof(double));
/* Compute X_diff, this is static for the entire algorithm. */
rowDiffs(X, n, p, X_diff);
for (attempt = 0; attempt < attempts; ++attempt) {
/* (Re)set learning rate. */
tau = tau_init;
/* Sample start value from stiefl manifold. */
rStiefl(p, q, V, workMem, workLen);
/* Create projection matrix for initial `V`. */
nullProj(V, p, q, Q);
/* Compute Distance vector. */
matrixprod(X, n, p, Q, p, p, X_proj); // here X_proj is only `(n, p)`.
rowDiffSquareSums(X_proj, n, p, vecD);
/* Apply kernel to distances. */
for (i = 0; i < nn; ++i) {
vecK[i] = gaussKernel(vecD[i], gKscale);
}
/* Compute col(row) sums of kernal vector (sym. packed lower tri
* matrix.), because `K == K^T` the rowSums are equal to colSums. */
rowSumsSymVec(vecK, n, 1.0, colSums);
/* Compute loss given the kernel vector and its column sums.
* Additionally the first momentum `y1` is computed and stored in
* the working memory (only intermidiate result, needed for `vecS`). */
loss_last = cost(n, Y, vecK, colSums, y1, L);
if (logger) {
callLogger(logger, loggerEnv,
attempt, 0,
L, n, V, p, q, tau);
}
/* Calc the scaling vector used for final computation of grad. */
scaling(n, Y, y1, L, vecD, vecK, colSums, vecS);
/* Compute the eucledian gradient `G`. */
rowSweep(X_diff, nn, p, "*", vecS, X_proj);
crossprod(X_diff, nn, p, X_proj, nn, p, workMem);
matrixprod(workMem, p, p, V, p, q, G);
scale(-2. / (((double)n) * h * h), G, p * q); // in-place
/* Compute Skew-Symmetric matrix `A` used in Cayley transform.
+ `A <- tau * (G V^T - V G^T) + 0 * A`*/
skew(p, q, tau, G, V, 0.0, A);
for (epoch = 0; epoch < epochs; ++epoch) {
/* Move V allong A */
cayleyTransform(p, q, A, V, V_tau, workMem);
/* Create projection matrix for `V_tau`. */
nullProj(V_tau, p, q, Q);
/* Compute Distance vector. */
matrixprod(X, n, p, Q, p, p, X_proj); // here X_proj only `(n, p)`.
rowDiffSquareSums(X_proj, n, p, vecD);
/* Apply kernel to distances. */
for (i = 0; i < nn; ++i) {
vecK[i] = gaussKernel(vecD[i], gKscale);
}
/* Compute col(row) sums of kernal vector (sym. packed lower tri
* matrix.), because `K == K^T` the rowSums are equal to colSums. */
rowSumsSymVec(vecK, n, 1.0, colSums);
/* Compute loss given the kernel vector and its column sums.
* Additionally the first momentum `y1` is computed and stored in
* the working memory (only intermidiate result, needed for `vecS`). */
loss = cost(n, Y, vecK, colSums, y1, L);
/* Check if step is appropriate, iff not reduce learning rate. */
if ((loss - loss_last) > 0.0) {
tau *= 0.5;
scale(0.5, A, p * p);
continue;
}
// Compute error, use workMem (keep first `n`, they store `y1`).
skew(p, q, 1.0, V, V_tau, 0.0, workMem);
err = norm(workMem, p, p, "F");
// Shift next step to current step and store loss to last.
memcpy(V, V_tau, p * q * sizeof(double));
loss_last = loss;
if (logger) {
callLogger(logger, loggerEnv,
attempt, epoch + 1,
L, n, V, p, q, tau);
}
// Check Break condition.
if (err < tol || epoch + 1 >= epochs) {
break;
}
/* Continue computing the gradient. */
/* Calc the scaling vector used for final computation of grad. */
scaling(n, Y, y1, L, vecD, vecK, colSums, vecS);
/* Compute the eucledian gradient `G`. */
rowSweep(X_diff, nn, p, "*", vecS, X_proj);
crossprod(X_diff, nn, p, X_proj, nn, p, workMem);
matrixprod(workMem, p, p, V, p, q, G);
scale(-2. / (((double)n) * h * h), G, p * q); // in-place
/* Compute Skew-Symmetric matrix `A` used in Cayley transform.
+ `A <- tau * (G V^T - V G^T) + 0 * A`*/
skew(p, q, tau, G, V, 0.0, A);
}
/* Check if current attempt improved previous ones */
if (attempt == 0 || loss < loss_best) {
loss_best = loss;
memcpy(V_best, V, p * q * sizeof(double));
memcpy(L_best, L, n * sizeof(double));
}
}
memcpy(V, V_best, p * q * sizeof(double));
memcpy(L, L_best, n * sizeof(double));
}

View File

@ -0,0 +1,73 @@
#include "cve.h"
int getWorkLen(const int n, const int p, const int q) {
int mpq; /**< Max of p and q */
int nn = ((n - 1) * n) / 2;
if (p > q) {
mpq = p;
} else {
mpq = q;
}
if (nn * p < (mpq + 1) * mpq) {
return 2 * (mpq + 1) * mpq;
} else {
return (nn + mpq) * mpq;
}
}
double cost(const int n,
const double *Y,
const double *vecK,
const double *colSums,
double *y1, double *L) {
int i, j, k;
double tmp;
for (i = 0; i < n; ++i) {
y1[i] = Y[i];
L[i] = Y[i] * Y[i];
}
for (k = j = 0; j < n; ++j) {
for (i = j + 1; i < n; ++i, ++k) {
y1[i] += Y[j] * vecK[k];
y1[j] += Y[i] * vecK[k];
L[i] += Y[j] * Y[j] * vecK[k];
L[j] += Y[i] * Y[i] * vecK[k];
}
}
for (i = 0; i < n; ++i) {
y1[i] /= colSums[i];
L[i] /= colSums[i];
}
tmp = 0.0;
for (i = 0; i < n; ++i) {
tmp += (L[i] -= y1[i] * y1[i]);
}
return tmp / (double)n;
}
void scaling(const int n,
const double *Y, const double *y1, const double *L,
const double *vecD, const double *vecK,
const double *colSums,
double *vecS) {
int i, j, k, nn = (n * (n - 1)) / 2;
double tmp;
for (k = j = 0; j < n; ++j) {
for (i = j + 1; i < n; ++i, ++k) {
tmp = Y[j] - y1[i];
vecS[k] = (L[i] - (tmp * tmp)) / colSums[i];
tmp = Y[i] - y1[j];
vecS[k] += (L[j] - (tmp * tmp)) / colSums[j];
}
}
for (k = 0; k < nn; ++k) {
vecS[k] *= vecK[k] * vecD[k];
}
}

View File

@ -1,28 +1,56 @@
#include <Rinternals.h>
#include "cve.h"
void grad(const int n, const int p, const int q,
const double *X,
const double *X_diff,
const double *Y,
const double *V,
const double h,
double *G, double *loss);
// SEXP rStiefl_c(SEXP pin, SEXP qin) {
// int p = asInteger(pin);
// int q = asInteger(qin);
SEXP grad_c(SEXP X, SEXP X_diff, SEXP Y, SEXP V, SEXP h) {
SEXP G = PROTECT(allocMatrix(REALSXP, nrows(V), ncols(V)));
SEXP loss = PROTECT(ScalarReal(0.0));
// SEXP Vout = PROTECT(allocMatrix(REALSXP, p, q));
grad(nrows(X), ncols(X), ncols(V),
REAL(X), REAL(X_diff), REAL(Y), REAL(V), *REAL(h),
REAL(G), REAL(loss));
// int workLen = 2 * (p + 1) * q;
// double *workMem = (double*)R_alloc(workLen, sizeof(double));
// rStiefl(p, q, REAL(Vout), workMem, workLen);
// UNPROTECT(1);
// return Vout;
// }
SEXP cve_simple(SEXP X, SEXP Y, SEXP k, SEXP h,
SEXP tau, SEXP tol,
SEXP epochs, SEXP attempts,
SEXP logger, SEXP loggerEnv) {
/* Handle logger parameter, set to NULL pointer if not a function. */
if (!(isFunction(logger) && isEnvironment(loggerEnv))) {
logger = (SEXP)0;
}
/* Get dimensions. */
int n = nrows(X);
int p = ncols(X);
int q = p - asInteger(k);
/* Convert types if needed. */
// TODO:
/* Create output list. */
SEXP Vout = PROTECT(allocMatrix(REALSXP, p, q));
SEXP Lout = PROTECT(allocVector(REALSXP, n));
/* Call CVE simple subroutine. */
cve_simple_sub(n, p, q,
REAL(X), REAL(Y), asReal(h),
asReal(tau), asReal(tol),
asInteger(epochs), asInteger(attempts),
REAL(Vout), REAL(Lout),
logger, loggerEnv);
SEXP out = PROTECT(allocVector(VECSXP, 2));
SET_VECTOR_ELT(out, 0, G);
SET_VECTOR_ELT(out, 1, loss);
SET_VECTOR_ELT(out, 0, Vout);
SET_VECTOR_ELT(out, 1, Lout);
SEXP names = PROTECT(allocVector(STRSXP, 2));
SET_STRING_ELT(names, 0, mkChar("G"));
SET_STRING_ELT(names, 1, mkChar("loss"));
setAttrib(out, install("names"), names);
SET_STRING_ELT(names, 0, mkChar("V"));
SET_STRING_ELT(names, 1, mkChar("L"));
setAttrib(out, R_NamesSymbol, names);
UNPROTECT(4);
return out;

View File

@ -1,123 +0,0 @@
#include <stdlib.h>
#include <math.h>
#include "sums.h"
#include "sweep.h"
#include "matrix.h"
#include "indexing.h"
// TODO: clarify
static inline double gaussKernel(const double x, const double scale) {
return exp(scale * x * x);
}
// TODO: mutch potential for optimization!!!
static void weightedYandLoss(const int n,
const double *Y,
const double *vecD,
const double *vecW,
const double *colSums,
double *y1, double *L, double *vecS,
double *loss) {
int i, j, k, N = n * (n - 1) / 2;
double l;
for (i = 0; i < n; ++i) {
y1[i] = Y[i] / colSums[i];
L[i] = Y[i] * Y[i] / colSums[i];
}
for (k = j = 0; j < n; ++j) {
for (i = j + 1; i < n; ++i, ++k) {
y1[i] += Y[j] * vecW[k] / colSums[i];
y1[j] += Y[i] * vecW[k] / colSums[j];
L[i] += Y[j] * Y[j] * vecW[k] / colSums[i];
L[j] += Y[i] * Y[i] * vecW[k] / colSums[j];
}
}
l = 0.0;
for (i = 0; i < n; ++i) {
l += (L[i] -= y1[i] * y1[i]);
}
*loss = l / (double)n;
for (k = j = 0; j < n; ++j) {
for (i = j + 1; i < n; ++i, ++k) {
l = Y[j] - y1[i];
vecS[k] = (L[i] - (l * l)) / colSums[i];
l = Y[i] - y1[j];
vecS[k] += (L[j] - (l * l)) / colSums[j];
}
}
for (k = 0; k < N; ++k) {
vecS[k] *= vecW[k] * vecD[k];
}
}
void grad(const int n, const int p, const int q,
const double *X,
const double *X_diff,
const double *Y,
const double *V,
const double h,
double *G, double *loss) {
// Number of X_i to X_j not trivial pairs.
int i, N = (n * (n - 1)) / 2;
double scale = -0.5 / h;
if (X_diff == (void*)0) {
// TODO: ...
}
// Allocate and compute projection matrix `Q = I_p - V * V^T`
double *Q = (double*)malloc(p * p * sizeof(double));
nullProj(V, p, q, Q);
// allocate and compute vectorized distance matrix with a temporary
// projection of `X_diff`.
double *vecD = (double*)malloc(N * sizeof(double));
double *X_proj;
if (p < 5) { // TODO: refine that!
X_proj = (double*)malloc(N * 5 * sizeof(double));
} else {
X_proj = (double*)malloc(N * p * sizeof(double));
}
matrixprod(X_diff, N, p, Q, p, p, X_proj);
rowSquareSums(X_proj, N, p, vecD);
// Apply kernel to distence vector for weights computation.
double *vecW = X_proj; // reuse memory area, no longer needed.
for (i = 0; i < N; ++i) {
vecW[i] = gaussKernel(vecD[i], scale);
}
double *colSums = X_proj + N; // still allocated!
rowSumsSymVec(vecW, n, 1.0, colSums); // rowSums = colSums cause Sym
// compute weighted responces of first end second momontum, aka y1, y2.
double *y1 = X_proj + N + n;
double *L = X_proj + N + (2 * n);
// Allocate X_diff scaling vector `vecS`, not in `X_proj` mem area because
// used symultanious to X_proj in final gradient computation.
double *vecS = (double*)malloc(N * sizeof(double));
weightedYandLoss(n, Y, vecD, vecW, colSums, y1, L, vecS, loss);
// compute the gradient using X_proj for intermidiate scaled X_diff.
rowSweep(X_diff, N, p, "*", vecS, X_proj);
// reuse Q which has the required dim (p, p).
crossprod(X_diff, N, p, X_proj, N, p, Q);
// Product with V
matrixprod(Q, p, p, V, p, q, G);
// And final scaling (TODO: move into matrixprod!)
scale = -2.0 / (((double)n) * h * h);
N = p * q;
for (i = 0; i < N; ++i) {
G[i] *= scale;
}
free(vecS);
free(X_proj);
free(vecD);
free(Q);
}

View File

@ -1,12 +0,0 @@
#include "indexing.h"
void rangePairs(const int from, const int to, int *pairs) {
int i, j, k;
for (k = 0, i = from; i < to; ++i) {
for (j = i + 1; j < to; ++j, k += 2) {
pairs[k] = i;
pairs[k + 1] = j;
}
}
}

View File

@ -1,8 +0,0 @@
/* Include Guard */
#ifndef CVE_INCLUDE_GUARD_INDEXING_
#define CVE_INCLUDE_GUARD_INDEXING_
void rangePairs(const int from, const int to, int *pairs);
#endif /* CVE_INCLUDE_GUARD_INDEXING_ */

View File

@ -8,10 +8,14 @@
*/
/* .Call calls */
extern SEXP grad_c(SEXP, SEXP, SEXP, SEXP, SEXP);
extern SEXP cve_simple(SEXP X, SEXP Y, SEXP k,
SEXP h,
SEXP tau, SEXP tol,
SEXP epochs, SEXP attempts,
SEXP logger, SEXP loggerEnv);
static const R_CallMethodDef CallEntries[] = {
{"grad_c", (DL_FUNC) &grad_c, 5},
{"cve_simple", (DL_FUNC) &cve_simple, 10},
{NULL, NULL, 0}
};

View File

@ -1,9 +1,26 @@
#include <string.h> // for `mem*` functions.
#include "cve.h"
#include "config.h"
#include "matrix.h"
double norm(const double *A, const int nrow, const int ncol,
const char *type) {
int i, nelem = nrow * ncol;
int nelemb = (nelem / 4) * 4;
double sum = 0.0;
#include <R_ext/BLAS.h>
if (*type == 'F') {
for (i = 0; i < nelemb; i += 4) {
sum += A[i] * A[i]
+ A[i + 1] * A[i + 1]
+ A[i + 2] * A[i + 2]
+ A[i + 3] * A[i + 3];
}
for (; i < nelem; ++i) {
sum += A[i] * A[i];
}
return sqrt(sum);
} else {
error("Unknown norm type.");
}
}
void matrixprod(const double *A, const int nrowA, const int ncolA,
const double *B, const int nrowB, const int ncolB,
@ -35,12 +52,12 @@ void nullProj(const double *B, const int nrowB, const int ncolB,
double *Q) {
const double minusOne = -1.0;
const double one = 1.0;
int i, nelem = nrowB * nrowB;
// Initialize as identity matrix.
memset(Q, 0, sizeof(double) * nrowB * nrowB);
double *Q_diag, *Q_end = Q + nrowB * nrowB;
for (Q_diag = Q; Q_diag < Q_end; Q_diag += nrowB + 1) {
*Q_diag = 1.0;
for (i = 0; i < nelem; i += nrowB + 1) {
Q[i] = 1.0;
}
// DGEMM with parameterization:
@ -50,13 +67,32 @@ void nullProj(const double *B, const int nrowB, const int ncolB,
&one, Q, &nrowB);
}
// In place scaling of elements of A.
void scale(const double s, double *A, const int nelem) {
int i, nelemb = (nelem / 4) * 4;
if (s == 1.) {
return;
}
for (i = 0; i < nelemb; i += 4) {
A[i] *= s;
A[i + 1] *= s;
A[i + 2] *= s;
A[i + 3] *= s;
}
for (; i < nelem; ++i) {
A[i] *= s;
}
}
// A dence skwe-symmetric rank 2 update.
// Perform the update
// C := alpha (A * B^T - B * A^T) + beta C
void skewSymRank2k(const int nrow, const int ncol,
double alpha, const double *A, const double *B,
double beta,
double *C) {
void skew(const int nrow, const int ncol,
double alpha, const double *A, const double *B,
double beta,
double *C) {
F77_NAME(dgemm)("N", "T",
&nrow, &nrow, &ncol,
&alpha, A, &nrow, B, &nrow,
@ -69,3 +105,60 @@ void skewSymRank2k(const int nrow, const int ncol,
&alpha, B, &nrow, A, &nrow,
&beta, C, &nrow);
}
/** Cayley transformation of matrix `B` using the Skew-Symmetri matrix `A`.
* X = (I + A)^-1 (I - A) B
* by solving the following linear equation:
* (I + A) X = (I - A) B ==> X = (I + A)^-1 (I - A) B
* \_____/ \_____/
* IpA X = ImA B
* \_______/
* IpA X = Y ==> X = IpA^-1 Y
*
* @param A Skew-Symmetric matrix of dimension `(n, n)`.
* @param B Matrix of dimensions `(n, m)` with `m <= n`.
* @return Transformed matrix `X`.
* @note This opperation is equivalent to the R expression:
* solve(diag(1, n) + A) %*% (diag(1, n) - A) %*% B
* or
* solve(diag(1, n) + A, (diag(1, n) - A) %*% B)
*/
void cayleyTransform(const int p, const int q,
const double *A, const double *B,
double *X, double *workMem) {
int i, info, pp = p * p;
double zero = 0., one = 1.;
/* Allocate row permutation array used by `dgesv` */
int *ipiv = (int*)workMem;
/* Create Matrix IpA = I + A (I plus A) */
double *IpA = (double*)(ipiv + p);
memcpy(IpA, A, pp * sizeof(double));
for (i = 0; i < pp; i += p + 1) {
IpA[i] += 1.; // +1 to diagonal elements.
}
/* Create Matrix ImA = I - A (I minus A) */
double *ImA = IpA + pp;
for (i = 0; i < pp; ++i) {
ImA[i] = -A[i];
}
for (i = 0; i < pp; i += p + 1) {
ImA[i] += 1.; // +1 to diagonal elements.
}
/* Y as matrix-matrix product of ImA and B:
* Y = 1 * ImA * B + 0 * Y */
F77_CALL(dgemm)("N", "N", &p, &q, &p,
&one, ImA, &p, B, &p, &zero, X, &p);
/* Solve system IpA Y = X for Y (and store result in X).
* aka. X = IpA^-1 X */
F77_CALL(dgesv)(&p, &q, IpA, &p, ipiv, X, &p, &info);
if (info) {
error("[ cayleyTransform ] error in dgesv - info %d", info);
}
}

View File

@ -1,25 +0,0 @@
/* Include Guard */
#ifndef CVE_INCLUDE_GUARD_MATRIX_
#define CVE_INCLUDE_GUARD_MATRIX_
void matrixprod(const double *A, const int nrowA, const int ncolA,
const double *B, const int nrowB, const int ncolB,
double *C);
void crossprod(const double *A, const int nrowA, const int ncolA,
const double *B, const int nrowB, const int ncolB,
double *C);
void nullProj(const double *B, const int nrowB, const int ncolB,
double *Q);
// A dence skwe-symmetric rank 2 update.
// Perform the update
// C := alpha (A * B^T - B * A^T) + beta C
void skewSymRank2k(const int nrow, const int ncol,
double alpha, const double *A, const double *B,
double beta,
double *C);
#endif /* CVE_INCLUDE_GUARD_MATRIX_ */

81
CVE_C/src/rStiefl.c Normal file
View File

@ -0,0 +1,81 @@
#include "cve.h"
// /**
// * Performas a QR factorization and computes the Q factor.
// *
// * @param A matrix.
// * @returns The Q factor of the QR factorization `A = QR`.
// */
// SEXP qrQ(SEXP Ain) {
// int i, j, info;
// if (!isMatrix(Ain)) {
// error("Argument must be a matrix.");
// }
// int nrow = nrows(Ain);
// int ncol = ncols(Ain);
// double *A = (double*)R_alloc(nrow * ncol, sizeof(double));
// memcpy(A, REAL(Ain), nrow * ncol * sizeof(double));
// // double *A = REAL(Ain);
// // Scalar factors of elementary reflectors.
// double *tau = (double*)R_alloc(ncol, sizeof(double));
// // Create Working memory area.
// int lenWork = 3 * nrow;
// double *memWork = (double*)R_alloc(lenWork, sizeof(double));
// F77_NAME(dgeqrf)(&nrow, &ncol, A, &nrow, tau,
// memWork, &lenWork, &info);
// SEXP Qout = PROTECT(allocMatrix(REALSXP, nrow, ncol));
// double *Q = REAL(Qout);
// for (j = 0; j < ncol; ++j) {
// for (i = 0; i < nrow; ++i) {
// if (i == j) {
// Q[i + nrow * j] = 1.;
// } else {
// Q[i + nrow * j] = 0.;
// }
// }
// }
// F77_NAME(dormqr)("L", "N", &nrow, &ncol, &ncol, A, &nrow, tau, Q, &nrow,
// memWork, &lenWork, &info);
// UNPROTECT(1);
// return Qout;
// }
void rStiefl(const int p, const int q, double *V,
double *workMem, int workLen) {
int i, j, info;
int pq = p * q;
GetRNGstate();
for (i = 0; i < pq; ++i) {
workMem[i] = norm_rand();
}
PutRNGstate();
double *tau = workMem + pq;
workLen -= pq + q;
F77_CALL(dgeqrf)(&p, &q, workMem, &p, tau,
workMem + pq + q, &workLen, &info);
for (j = 0; j < q; ++j) {
for (i = 0; i < p; ++i) {
if (i == j) {
V[i + p * j] = 1.;
} else {
V[i + p * j] = 0.;
}
}
}
F77_NAME(dormqr)("L", "N", &p, &q, &q, workMem, &p, tau, V, &p,
workMem + pq + q, &workLen, &info);
}

View File

@ -1,8 +1,12 @@
#include <string.h> // for `mem*` functions.
#include "config.h"
#include "sums.h"
#include "cve.h"
/**
* Computes the row sums of a matrix `A`.
* @param A Pointer to col-major matrix elements, size is `nrow * ncol`.
* @param nrow Number of rows of `A`.
* @param ncol Number of columns of `A`.
* @param sum Pointer to output row sums of size `nrow`.
*/
void rowSums(const double *A, const int nrow, const int ncol,
double *sum) {
int i, j, block_size, block_size_i;
@ -41,29 +45,38 @@ void rowSums(const double *A, const int nrow, const int ncol,
}
void colSums(const double *A, const int nrow, const int ncol,
double *sum) {
int j;
double *sum_end = sum + ncol;
double *colSums) {
int i, j;
int nrowb = 4 * (nrow / 4); // 4 * floor(nrow / 4)
double colSum;
memset(sum, 0, sizeof(double) * ncol);
for (; sum < sum_end; ++sum) {
for (j = 0; j < nrow; ++j) {
*sum += A[j];
for (j = 0; j < ncol; ++j) {
colSum = 0.0;
for (i = 0; i < nrowb; i += 4) {
colSum += A[i]
+ A[i + 1]
+ A[i + 2]
+ A[i + 3];
}
for (; i < nrow; ++i) {
colSum += A[i];
}
*(colSums++) = colSum;
A += nrow;
}
}
void rowSquareSums(const double *A, const int nrow, const int ncol,
void rowSquareSums(const double *A,
const int nrow, const int ncol,
double *sum) {
int i, j, block_size, block_size_i;
const double *A_block = A;
const double *A_end = A + nrow * ncol;
if (nrow < CVE_MEM_CHUNK_SIZE) {
block_size = nrow;
} else {
if (nrow > CVE_MEM_CHUNK_SIZE) {
block_size = CVE_MEM_CHUNK_SIZE;
} else {
block_size = nrow;
}
// Iterate `(block_size_i, ncol)` submatrix blocks.
@ -111,3 +124,37 @@ void rowSumsSymVec(const double *Avec, const int nrow,
}
}
}
void rowDiffs(const double* X, const int nrow, const int ncol,
double *diffs) {
int i, j, k, l;
const double *Xcol;
for (k = l = 0; l < ncol; ++l) {
Xcol = X + l * nrow;
for (i = 0; i < nrow; ++i) {
for (j = i + 1; j < nrow; ++j) {
diffs[k++] = Xcol[i] - Xcol[j];
}
}
}
}
void rowDiffSquareSums(const double* X, const int nrow, const int ncol,
double *sum) {
int i, j, k, l;
const double *Xcol;
double tmp;
memset(sum, 0, ((nrow * (nrow - 1)) / 2) * sizeof(double));
for (l = 0; l < ncol; ++l) {
Xcol = X + l * nrow;
for (k = i = 0; i < nrow; ++i) {
for (j = i + 1; j < nrow; ++j, ++k) {
tmp = Xcol[i] - Xcol[j];
sum[k] += tmp * tmp;
}
}
}
}

View File

@ -1,19 +0,0 @@
/* Include Guard */
#ifndef CVE_INCLUDE_GUARD_SUMS_
#define CVE_INCLUDE_GUARD_SUMS_
void rowSums(const double *A, const int nrow, const int ncol,
double *sum);
void colSums(const double *A, const int nrow, const int ncol,
double *sum);
void rowSquareSums(const double *A, const int nrow, const int ncol,
double *sum);
void rowSumsSymVec(const double *Avec, const int nrow,
const double diag,
double *sum);
#endif /* CVE_INCLUDE_GUARD_SUMS_ */

View File

@ -1,7 +1,4 @@
#include <R_ext/Error.h> // for `error`.
#include "config.h"
#include "sweep.h"
#include "cve.h"
/* C[, j] = A[, j] * v for each j = 1 to ncol */
void rowSweep(const double *A, const int nrow, const int ncol,
@ -107,7 +104,5 @@ void rowSweep(const double *A, const int nrow, const int ncol,
C_block += block_size_i;
v += block_size_i;
}
} else {
error("Got unknown 'op' (opperation) argument.");
}
}

View File

@ -1,11 +0,0 @@
/* Include Guard */
#ifndef CVE_INCLUDE_GUARD_SWEEP_
#define CVE_INCLUDE_GUARD_SWEEP_
void rowSweep(const double *A, const int nrow, const int ncol,
const char* op,
const double *v, // vector of length nrow
double *C);
#endif /* CVE_INCLUDE_GUARD_SWEEP_ */

51
test.R
View File

@ -1,23 +1,35 @@
args <- commandArgs(TRUE)
if (length(args) > 0) {
method <- args[1]
} else {
method <- "simple"
}
epochs <- 50L
attempts <- 25L
# library(CVEpureR)
# path <- '~/Projects/CVE/tmp/logger.R.pdf'
# path <- paste0('~/Projects/CVE/tmp/logger_', method, '.R.pdf')
library(CVE)
path <- '~/Projects/CVE/tmp/logger.C.pdf'
path <- paste0('~/Projects/CVE/tmp/logger_', method, '.C.pdf')
epochs <- 100
attempts <- 25
# Define the logger for the `cve()` method.
logger <- function(env) {
# Define logger for `cve()` method.
logger <- function(epoch, attempt, L, V, tau) {
# Note the `<<-` assignement!
loss.history[env$epoch + 1, env$attempt] <<- env$loss
error.history[env$epoch + 1, env$attempt] <<- env$error
tau.history[env$epoch + 1, env$attempt] <<- env$tau
loss.history[epoch + 1, attempt] <<- mean(L)
if (epoch == 0) {
error <- NA
} else {
error <- norm(V %*% t(V) - V_last %*% t(V_last), type = 'F')
}
V_last <<- V
error.history[epoch + 1, attempt] <<- error
tau.history[epoch + 1, attempt] <<- tau
# Compute true error by comparing to the true `B`
B.est <- null(env$V) # Function provided by CVE
B.est <- null(V) # Function provided by CVE
P.est <- B.est %*% solve(t(B.est) %*% B.est) %*% t(B.est)
true.error <- norm(P - P.est, 'F') / sqrt(2 * k)
true.error.history[env$epoch + 1, env$attempt] <<- true.error
true.error.history[epoch + 1, attempt] <<- true.error
}
pdf(path)
@ -37,12 +49,15 @@ for (name in paste0("M", seq(5))) {
P <- B %*% solve(t(B) %*% B) %*% t(B)
# Setup histories.
V_last <- NULL
loss.history <- matrix(NA, epochs + 1, attempts)
error.history <- matrix(NA, epochs + 1, attempts)
tau.history <- matrix(NA, epochs + 1, attempts)
true.error.history <- matrix(NA, epochs + 1, attempts)
dr <- cve(Y ~ X, k = k, logger = logger, epochs = epochs, attempts = attempts)
dr <- cve(Y ~ X, k = k, method = method,
epochs = epochs, attempts = attempts,
logger = logger)
# Plot history's
matplot(loss.history, type = 'l', log = 'y', xlab = 'i (iteration)',
@ -50,11 +65,13 @@ for (name in paste0("M", seq(5))) {
ylab = expression(L(V[i])))
matplot(true.error.history, type = 'l', log = 'y', xlab = 'i (iteration)',
main = paste('true error', name),
ylab = expression(group('|', B * B^T - B[i] * B[i]^T, '|')[F] / sqrt(2 * k)))
ylab = expression(group('|', B*B^T - B[i]*B[i]^T, '|')[F] / sqrt(2*k)))
matplot(error.history, type = 'l', log = 'y', xlab = 'i (iteration)',
main = paste('error', name),
ylab = expression(group('|', V[i-1] * V[i-1]^T - V[i] * V[i]^T, '|')[F]))
ylab = expression(group('|', V[i-1]*V[i-1]^T - V[i]*V[i]^T, '|')[F]))
matplot(tau.history, type = 'l', log = 'y', xlab = 'i (iteration)',
main = paste('learning rate', name),
ylab = expression(tau[i]))
}
}
cat("Created plot:", path, "\n")