Make A tridiagonal, upper, non-unit matrix by clearing the strictly lower part of the matrix.

Returns the norm of a vector in S

Compute in-place A += alpha for all elements in the matrix if list of indexes is empty. Otherwise compute A[i] += alpha for indexes in column-major or

Generic matrix-matrix multiplication for block [R:E, S:L] with panel length P. if trans is NOTRANS then calculates C = alpha*A*B + beta*C; C is M*N,

Compute element-wise A = Exp(A).

Matrix-vector rank update A = A + alpha*X*Y.T A is M*N generic matrix, X is row or column vector of length M Y is row or column vector of legth N.

NewFCBig returns a newly created FCBig adjusted for the closed interval [lo, hi] or an Error if any. If hq == true then trade some generation time for

Next returns the first PRN after Pos.

Ordinal gives you the input number in a rank/ordinal format. Ordinal(3) -> 3rd

20 bit factor of M100000223

* Evaluates x := H(lambda^{1/2}) * x (inverse is 'N') x := H(lambda^{-1/2}) * x (inverse is 'I'). H is the Hessian of the log

Compute in-place remainder A[i,j] %= alpha

getWordProb returns P(W|C_j) -- the probability of seeing a particular word W in a document of this class.

Create new zero filled matrix.

Set global decomposition block size for blocked versions.

Compute a Givens rotation (c,s) which zeroes the vector (a,b)

Create new matrix initialized to value.

Append dimension key to dis.

Set dimensions. Does not affect element allocations. Optional 3rd argument sets also the row stride of data.

Set values for sub-matrix starting at (row, col). If row+mat.Rows() greater than A.Rows() or col+mat.Cols() greater than A.Cols() matrix A is not chan

Compute a Givens rotation (c,s) which zeroes the vector (a,b)

Make A tridiagonal, lower, unit matrix by clearing the strictly upper part of the matrix and setting diagonal elements to one.

Calculate X = alpha*X.

Helper function, make a pseudo-random Vector with dimension dim.

=========================== [ General Methods ] ============================ Get the dimension of the Vector.

Test for equality. Return true if for all i,j: all A[i,j] = B[i,j]

* Get a submatrix starting at i,j with rows rows and cols columns. Changes to the returned matrix show up in the original. */

DFS for a floating point valued ConstantF

* lag-1.go * * Copyright (C) 1996, 1997, 1998, 1999, 2000, 2007 Jim Davies, Brian Gough * Copyright (C) 2012, 2013 G.vd.Schoot * * This program

Learn will accept new training documents for supervised learning.

NewFC32 returns a newly created FC32 adjusted for the closed interval [lo, hi] or an Error if any. If hq == true then trade some generation time for i

A = alpha*A + beta*B A = alpha*A + beta*B.T if flags&TRANSB

* wmean.go * * Copyright (C) 1996, 1997, 1998, 1999, 2000, 2007 Jim Davies, Brian Gough * Copyright (C) 2012, 2013 G.vd.Schoot * * This program

this will compute sub-block matrix product: Cij += Aik * Bkj using successive inner vector product (DOT) function.

Scale a tile of M rows by N columns with leading index ldX.

* Partition A to 1 by 2 blocks. A --> AL | AR Parameter nb is initial block size for AL (pLEFT) or AR (pRIGHT). */

Makes a copy C of A and applies function fn to elements of then new copy C. If indexes is non empty then function is applied to all elements of C inde

Generic triangular matrix update; blocked S is the start column and row in C; E is the end column and row in C

Compute element-wise division A /= B. Return A. If A and B sizes do not match A is returned unaltered.

Returns (mnl, x0) where mln number of nonlinear inequality constraints and x0 is a point in the domain of f.

has.go

LogScores produces "log-likelihood"-like scores that can be used to classify documents into classes. The value of the score is proportional to the li

Next returns the first PRN after Pos.

NextPrime returns first prime > n and true if successful or an undefined value and false if there is no next prime in the uint32 limits. Typical run t

Counts the number of indexed Constant nodes

Map parameter value to name string that can be used when calling Fortran library functions.

* ProbablyPrimeUint32 returns true if n is prime or n is a pseudoprime to base a. It implements the Miller-Rabin primality test for one specific value

Compute in-place A *= alpha for all elements in the matrix if list of indexes is empty. Otherwise compute A[i] *= alpha for indexes in column-major or

* Set the element in the ith row and jth column to v. */

Generic update for triangular lower or upper matrix. C = beta*C + alpha*A*B flags has NOTRANS C = beta*C + alpha*A*B.T flags has TRANS

Copying 4 columns parallel distributes cache misses even more evenly.

solve: op(A)*X = alpha*B or X*op(A) = alpha*B; unblocked version

X = A*X; unblocked version

Scale a vector of N elements with incX interval.

* * Merge 1 by 1 block from 2 by 1 block. * * AT * Abkl <-- -- * AB * */