glm non_negative coef argument added to R

Author: Amy Wang

R/h2o-package/R/Algorithms.R

@@ -262,7 +262,7 @@ h2o.glm <- function(x, y, data, key = "", family, link, nfolds = 0, alpha = 0.5,
                     epsilon = 1e-4, standardize = TRUE, prior, variable_importances = FALSE, use_all_factor_levels = FALSE,
                     tweedie.p = ifelse(family == "tweedie", 1.5, as.numeric(NA)), iter.max = 100,
                     higher_accuracy = FALSE, lambda_search = FALSE, return_all_lambda = FALSE, max_predictors=-1,
-                    offset, has_intercept = TRUE) {
+                    offset, has_intercept = TRUE, non_negative = FALSE) {
 
   if(!is.character(key)) stop("key must be of class character")
   if(nchar(key) > 0 && regexpr("^[a-zA-Z_][a-zA-Z0-9_.]*$", key)[1] == -1)
@@ -274,6 +274,7 @@ h2o.glm <- function(x, y, data, key = "", family, link, nfolds = 0, alpha = 0.5,
   if(!is.numeric(alpha)) stop('alpha must be numeric')
   if( any(alpha < 0) ) stop('alpha must be >= 0')
   if(!is.logical(has_intercept)) stop('has_intercept must be logical')
+  if(!is.logical(non_negative)) stop('non_negative must be logical')
   if(missing(offset)) { offset <- "" }
   else {
     if(!is.numeric(offset) && !is.character(offset)) stop("offset must be either an index or column name")
@@ -329,7 +330,7 @@ h2o.glm <- function(x, y, data, key = "", family, link, nfolds = 0, alpha = 0.5,
                               lambda_search = as.numeric(lambda_search), tweedie_variance_power = tweedie.p,
                               max_predictors = max_predictors, variable_importances = as.numeric(variable_importances),
                               use_all_factor_levels = as.numeric(use_all_factor_levels), link = link, offset = offset,
-                              has_intercept = as.numeric(has_intercept))
+                              has_intercept = as.numeric(has_intercept), non_negative = as.numeric(non_negative))
     else if(family == "binomial") {
       if(missing(prior)) prior = -1
       res = .h2o.__remoteSend(data@h2o, .h2o.__PAGE_GLM2, source = data@key, destination_key = key,
@@ -340,8 +341,8 @@ h2o.glm <- function(x, y, data, key = "", family, link, nfolds = 0, alpha = 0.5,
                               lambda_search = as.numeric(lambda_search), prior = prior,
                               max_predictors = max_predictors, variable_importances = as.numeric(variable_importances),
                               use_all_factor_levels = as.numeric(use_all_factor_levels), link = link, offset = offset,
-                              has_intercept = as.numeric(has_intercept))
-    } else
+                              has_intercept = as.numeric(has_intercept), non_negative = as.numeric(non_negative))
+    } else {
       res = .h2o.__remoteSend(data@h2o, .h2o.__PAGE_GLM2, source = data@key, destination_key = key,
                               response = args$y, ignored_cols = paste(x_ignore, sep="", collapse=","), family = family,
                               n_folds = nfolds, alpha = alpha, nlambdas = nlambda, lambda_min_ratio = lambda.min.ratio,
@@ -350,24 +351,24 @@ h2o.glm <- function(x, y, data, key = "", family, link, nfolds = 0, alpha = 0.5,
                               lambda_search = as.numeric(lambda_search),
                               max_predictors = max_predictors, variable_importances = as.numeric(variable_importances),
                               use_all_factor_levels = as.numeric(use_all_factor_levels), link = link, offset = offset,
-                              has_intercept = as.numeric(has_intercept))
+                              has_intercept = as.numeric(has_intercept), non_negative = as.numeric(non_negative)) }
 
     params = list(x=args$x, y=args$y, family = .h2o.__getFamily(family, tweedie.var.p=tweedie.p), nfolds=nfolds,
                   alpha=alpha, nlambda=nlambda, lambda.min.ratio=lambda.min.ratio, lambda=lambda,
                   beta_epsilon=epsilon, standardize=standardize, max_predictors = max_predictors,
                   variable_importances = variable_importances, use_all_factor_levels = use_all_factor_levels, h2o = data@h2o,
-                  link = link, offset = offset, has_intercept = as.numeric(has_intercept))
+                  link = link, offset = offset, has_intercept = as.numeric(has_intercept), non_negative = as.numeric(non_negative))
     .h2o.__waitOnJob(data@h2o, res$job_key)
     .h2o.get.glm(data@h2o, as.character(res$destination_key), return_all_lambda)
   } else
     .h2o.glm2grid.internal(x_ignore, args$y, data, key, family, link,nfolds, alpha, nlambda, lambda.min.ratio, lambda, epsilon,
                            standardize, prior, tweedie.p, iter.max, higher_accuracy, lambda_search, return_all_lambda,
                            variable_importances = variable_importances, use_all_factor_levels = use_all_factor_levels,
-                           offset = offset, has_intercept = as.numeric(has_intercept))
+                           offset = offset, has_intercept = as.numeric(has_intercept), non_negative = as.numeric(non_negative))
 }
 
 .h2o.glm2grid.internal <- function(x_ignore, y, data, key, family, link, nfolds, alpha, nlambda, lambda.min.ratio, lambda, epsilon, standardize, prior, tweedie.p, iter.max, higher_accuracy, lambda_search, return_all_lambda,
-                                   variable_importances, use_all_factor_levels, offset, has_intercept) {
+                                   variable_importances, use_all_factor_levels, offset, has_intercept, non_negative) {
   if(family == "tweedie")
     res = .h2o.__remoteSend(data@h2o, .h2o.__PAGE_GLM2, source = data@key, destination_key = key, response = y,
                             ignored_cols = paste(x_ignore, sep="", collapse=","), family = family, n_folds = nfolds,
@@ -376,7 +377,7 @@ h2o.glm <- function(x, y, data, key = "", family, link, nfolds = 0, alpha = 0.5,
                             higher_accuracy = as.numeric(higher_accuracy), lambda_search = as.numeric(lambda_search),
                             tweedie_variance_power = tweedie.p, variable_importances = as.numeric(variable_importances), 
                             use_all_factor_levels = as.numeric(use_all_factor_levels), link = link, offset = offset,
-                            has_intercept = as.numeric(has_intercept))
+                            has_intercept = as.numeric(has_intercept), non_negative = as.numeric(non_negative))
   else if(family == "binomial") {
     if(missing(prior)) prior = -1
     res = .h2o.__remoteSend(data@h2o, .h2o.__PAGE_GLM2, source = data@key, destination_key = key, response = y,
@@ -385,7 +386,7 @@ h2o.glm <- function(x, y, data, key = "", family, link, nfolds = 0, alpha = 0.5,
                             beta_epsilon = epsilon, standardize = as.numeric(standardize), max_iter = iter.max,
                             higher_accuracy = as.numeric(higher_accuracy), lambda_search = as.numeric(lambda_search), prior = prior,
                             variable_importances = as.numeric(variable_importances), use_all_factor_levels = as.numeric(use_all_factor_levels),
-                            link = link, offset = offset, has_intercept = as.numeric(has_intercept))
+                            link = link, offset = offset, has_intercept = as.numeric(has_intercept), non_negative = as.numeric(non_negative))
   }
   else
     res = .h2o.__remoteSend(data@h2o, .h2o.__PAGE_GLM2, source = data@key, destination_key = key, response = y,
@@ -394,13 +395,13 @@ h2o.glm <- function(x, y, data, key = "", family, link, nfolds = 0, alpha = 0.5,
                             beta_epsilon = epsilon, standardize = as.numeric(standardize), max_iter = iter.max,
                             higher_accuracy = as.numeric(higher_accuracy), lambda_search = as.numeric(lambda_search),
                             variable_importances = as.numeric(variable_importances), use_all_factor_levels = as.numeric(use_all_factor_levels), 
-                            link = link, offset = offset, has_intercept = as.numeric(has_intercept))
+                            link = link, offset = offset, has_intercept = as.numeric(has_intercept), non_negative = as.numeric(non_negative))
 
   params = list(x=setdiff(colnames(data)[-(x_ignore+1)], y), y=y, family=.h2o.__getFamily(family, tweedie.var.p=tweedie.p), 
                 link = link, nfolds=nfolds, alpha=alpha, nlambda=nlambda,
                 lambda.min.ratio=lambda.min.ratio, lambda=lambda, beta_epsilon=epsilon, standardize=standardize,
                 variable_importances = variable_importances, use_all_factor_levels = use_all_factor_levels, h2o = data@h2o,
-                offset = offset, has_intercept = as.numeric(has_intercept))
+                offset = offset, has_intercept = as.numeric(has_intercept), non_negative = as.numeric(non_negative))
 
   .h2o.__waitOnJob(data@h2o, res$job_key)
   .h2o.get.glm.grid(data@h2o, as.character(res$destination_key), return_all_lambda, data)

R/h2o-package/man/h2o.glm.Rd

@@ -12,7 +12,7 @@ h2o.glm(x, y, data, key = "", family, link, nfolds = 0, alpha = 0.5, nlambda = -
   prior, variable_importances = FALSE, use_all_factor_levels = FALSE, tweedie.p =
   ifelse(family == 'tweedie', 1.5, as.numeric(NA)), iter.max = 100, 
   higher_accuracy = FALSE, lambda_search = FALSE, return_all_lambda = FALSE, 
-  max_predictors = -1, offset, has_intercept = TRUE)
+  max_predictors = -1, offset, has_intercept = TRUE, non_negative = FALSE)
 }
 \arguments{
   \item{x}{
@@ -88,6 +88,7 @@ tweedie: tweedie \cr
   }
   \item{offset}{(Optional) Column to be used as an offset, if you have one. }
   \item{has_intercept}{ A logical value indicating whether or not to include the intercept term. If there are factor columns in your model, then the intercept must be included. }
+  \item{non_negative} { A logical value indicating whether or not coefficients will be restricted to only non-negative values.}
 }
 \value{
 An object of class \code{\linkS4class{H2OGLMModel}} with slots key, data, model and xval. The slot model is a list of the following components: