R/cosine_sim.R

In Swarchal/phenoDist: Distance Measurements in Phenotypic Screening

#' Creates a cosine similarity matrix
#'
#' For a given matrix, will create a cosine similarity matrix. Used as an
#' internal function for \code{cosine_sim()}
#'
#' @param X matrix
#' @param ... optional arguments to sum (i.e na.rm = TRUE)
#'
#' @return C cosine similarity matrix
#'
#' @export
#'
#' @examples
#' # example data
#' mat <- iris[,1:4]
#' out <- cosine_sim_mat(mat)
#' image(out)


cosine_sim_mat <- function(X, ...){

	cos_sim <- function(ix){
	    A = X[ix[1],]
	    B = X[ix[2],]
	    sum(A * B) / sqrt(sum(A^2, ...) * sum(B^2, ...))
	}

	# if given a dataframe, will try and convert to a matrix
	if (is.data.frame(X)){
		X <- as.matrix(X)
	}

	# check input
	if (!is.matrix(X)) stop("X needs to be a matrix")
	if (!is.numeric(X)) stop("X needs to be numeric")


    n <- nrow(X) 
    cmb <- expand.grid(i = 1:n, j = 1:n) 
    matrix(apply(cmb, 1, cos_sim), n, n)
}


#' Cosine similarity between two vectors
#'
#' Given two vectors, this function will calculate a cosine similarity.
#' Used as an internal function for \code{cosine_sim()}
#'
#' @param a vector
#' @param b vector
#' @param ... optional arguments to sum (i.e na.rm = TRUE)
#'
#' @return out cosine similarity
#'
#' @export
#'
#' @examples
#' set.seed(12321)
#' x <- rnorm(20)
#' y <- rnorm(20)
#' cosine_sim_vector(x, y)

cosine_sim_vector<- function(a, b, ...){

    if (is.data.frame(a) && nrow(a) == 1){
	    a <- as.numeric(a)
    }
    if (is.data.frame(b) && nrow(b) == 1){
	    b <- as.numeric(b)
    }

    # check inputs
    if (!is.vector(a) || !is.vector(b)){
	    stop("Inputs need to be vectors")
    }
    if (!is.numeric(a) || !is.numeric(b)){
	    stop("Inputs need to be numeric")
    }
    if (length(a) != length(b)){
	    stop("'a' and 'b' need to be the same length")
    }

    sum(a * b, ...) / sqrt(sum(a^2, ...) * sum(b^2, ...))
}





#' Cosine similarity 
#' 
#' Calculates a cosine similarity of either a matrix or between two vectors.
#' If given a matrix, will return a cosine-similarity matrix. If given two
#' vectors will calculate the cosine-similarity between the two. Specific
#' functions for cosine vectors or matrices can be called with
#' \code{cosine_sim_vector} and \code{cosine_sim_mat} respectively.
#'
#' @param ... Either a numerical matrix, or two numerical vectors of the same
#' 		length.
#' 
#' @return out A matrix if given a matrix, or a single number if given two
#' 		vectors.
#'
#' @export
#'
#' @examples
#' # example matrix
#' mat <- as.matrix(iris[,1:4])
#' out <- cosine_sim(mat)
#' image(out)
#'
#' # two vectors
#' a <- rnorm(20)
#' b <- rnorm(20)
#' cosine_sim(a, b)

cosine_sim <- function(...){

    dots <- list(...)
    if (length(dots) == 1){
        # a matrix
        cosine_sim_mat(dots[[1]])
    } else if (length(dots) == 2){
    	# two vectors
        cosine_sim_vector(dots[[1]], dots[[2]])
    } else stop("Need either a matrix or two vectors")
}