Merge branch 'master' of https://github.com/ksipos/polysemous_words

antoine/heatmap.R

@@ -8,6 +8,9 @@ use_condaenv('my_env_3') # just a Python 3 environment with the 'rbo' module ins
 
 rbo = import('rbo') # see https://github.com/changyaochen/rbo
 
+method_names_en = c('oxford','ontonotes','wikipedia','wndomains','wordnet_original','wordnet_restricted')
+method_names_fr = c('larousse','wikipedia')
+
 # = = = = = = = = = = = = = = = = functions
 
 dcg = function(x) {
@@ -76,8 +79,6 @@ score_ranking = function(evaluated,gt,metric){
 
 # = = = = = = = = = = = = = = = = arguments
 
-# TODO heatmap of score vs PCA dimensions and max levels
-
 args = commandArgs(trailingOnly=TRUE)
 
 # Rscript --vanilla heatmap.R path_root metric
@@ -105,6 +106,14 @@ if (!language%in%c('english','french')){
  stop(paste(language,'is currently not supported!'))
 }
 
+if (language=='english'){
+ method_names = method_names_en
+}
+
+if (language=='french'){
+ method_names = method_names_fr
+}
+
 # relevant links about the metrics:
 # - https://en.wikipedia.org/wiki/Learning_to_rank#Evaluation_measures
 # see: https://stats.stackexchange.com/questions/8071/how-to-choose-between-pearson-and-spearman-correlation
@@ -287,7 +296,8 @@ rankings[['random']] = lapply(1:n_runs,function(x){
 })
 
 # re-order/re-name to optimize the heatmap (our method, random, and frequency first)
-method_names = c(best_name_renamed,'random','frequency','oxford','ontonotes','wikipedia','wndomains','wordnet_original','wordnet_restricted')
+method_names = c(c(best_name_renamed,'random','frequency'),method_names)
+
 method_names_pretty = gsub('_',' ',method_names)
 
 pdf(paste0(path_to_plots,'score_distributions.pdf'),paper='a4r',width=10,height=7)

antoine/pyramid_scores_by_hand.R

@@ -1,15 +1,14 @@
-my_score = function(level,n_levels,coverage){
- coverage/(2^(n_levels-level))
-}
+levels = 0:2
 
-levels = 1:3
+n_dims = 2
+n_bins = unlist(lapply(levels,function(x) 2^(x*n_dims)))
 
 # word 1
-covs = c(3/4,7/16,10/64)
-sum(unlist(lapply(levels,function(level) my_score(level,length(levels),covs[level]))))
+covs = c(3,7,10)/n_bins
+sum(unlist(lapply(levels,function(level) covs[level+1]/(2^(max(levels)-level)))))
 
 # word 2
-covs = c(1/4,4/16,7/64)
-sum(unlist(lapply(levels,function(level) my_score(level,length(levels),covs[level]))))
+covs = c(1,4,7)/n_bins
+sum(unlist(lapply(levels,function(level) covs[level+1]/(2^(max(levels)-level)))))