query-simil.R

#' Calculate similarities from keys to keys
#'
#' @param conn a Magnitude connection.
#' @param keys character vector.
#' @param q character vector.
#' @param normalized logical;
#' whether or not vector embeddings should be normalized?
#' @param method string; method to compute similarity.
#' @param ... other arguments are passed to \code{proxyC::simil}.
#' @returns a sparse Matrix of 'Matrix' package.
#' @export
calc_simil <- function(conn, keys, q, normalized = TRUE,
                       method = c(
                         "cosine",
                         "correlation",
                         "jaccard",
                         "ejaccard",
                         "dice",
                         "edice",
                         "hamann",
                         "simple matching",
                         "faith"
                       ),
                       ...) {
  method <- rlang::arg_match(method)
  x <- query(conn, keys, normalized) %>%
    tibble::column_to_rownames("key") %>%
    as.matrix()
  y <- query(conn, q, normalized) %>%
    tibble::column_to_rownames("key") %>%
    as.matrix()
  proxyC::simil(x, y, method = method, ...)
}

#' Order keys by their similarity to a key
#'
#' @param conn a Magnitude connection.
#' @param key string.
#' @param q character vector.
#' elements exact same with key will be dropped from result.
#' @param n integer.
#' @param normalized logical;
#' whether or not vector embeddings should be normalized?
#' @param method string; method to compute similarity.
#' @returns a tibble.
#' @export
most_similar <- function(conn, key, q, n = 1L,
                         normalized = TRUE,
                         method = c(
                           "cosine",
                           "correlation",
                           "jaccard",
                           "ejaccard",
                           "dice",
                           "edice",
                           "hamann",
                           "simple matching",
                           "faith"
                         )) {
  if (length(key) != 1L) {
    rlang::warn("length of `key` is not 1L. the first element will be used.")
  }
  q <- unique(q[which(!q %in% key, arr.ind = TRUE)])
  n <- ifelse(n > length(q), length(q), n)
  simil <-
    as.matrix(calc_simil(conn, key[1], q, normalized, method))
  tibble::tibble(
    keys = colnames(simil),
    similarity = simil[1, ]
  ) %>%
    dplyr::arrange(desc(.data$similarity)) %>%
    dplyr::slice_head(n = n)
}