Octave can be used in demo in lieu of matlab.

demo.sh

@@ -37,6 +37,8 @@ if [[ $? -eq 0 ]]
  then
  if [ "$1" = 'matlab' ]; then
  matlab -nodisplay -nodesktop -nojvm -nosplash < ./eval/matlab/read_and_evaluate.m 1>&2 
+ elif [ "$1" = 'octave' ]; then
+ octave < ./eval/octave/read_and_evaluate_octave.m 1>&2 
  else
  python eval/python/evaluate.py
  fi

eval/octave/WordLookup_octave.m

@@ -0,0 +1,10 @@
+function index = WordLookup_octave(InputString)
+global wordMap
+
+if isfield(wordMap, InputString)
+ index = wordMap.(InputString);
+elseif isfield(wordMap, '<unk>')
+ index = wordMap.('<unk>');
+else
+ index = 0;
+end

eval/octave/evaluate_vectors_octave.m

@@ -0,0 +1,80 @@
+function [BB] = evaluate_vectors_octave(W)
+
+global wordMap
+
+filenames = {'capital-common-countries' 'capital-world' 'currency' 'city-in-state' 'family' 'gram1-adjective-to-adverb' ...
+ 'gram2-opposite' 'gram3-comparative' 'gram4-superlative' 'gram5-present-participle' 'gram6-nationality-adjective' ...
+ 'gram7-past-tense' 'gram8-plural' 'gram9-plural-verbs'};
+path = './eval/question-data/';
+
+split_size = 100; %to avoid memory overflow, could be increased/decreased depending on system and vocab size
+
+correct_sem = 0; %count correct semantic questions
+correct_syn = 0; %count correct syntactic questions
+correct_tot = 0; %count correct questions
+count_sem = 0; %count all semantic questions
+count_syn = 0; %count all syntactic questions
+count_tot = 0; %count all questions
+full_count = 0; %count all questions, including those with unknown words
+
+
+if isfield(wordMap, '<unk>')
+ unkkey = wordMap.('<unk>');
+else
+ unkkey = 0;
+end
+
+for j=1:length(filenames);
+
+clear dist;
+
+fid=fopen([path filenames{j} '.txt']);
+temp=textscan(fid,'%s%s%s%s');
+fclose(fid);
+ind1 = cellfun(@WordLookup_octave,temp{1}); %indices of first word in analogy
+ind2 = cellfun(@WordLookup_octave,temp{2}); %indices of second word in analogy
+ind3 = cellfun(@WordLookup_octave,temp{3}); %indices of third word in analogy
+ind4 = cellfun(@WordLookup_octave,temp{4}); %indices of answer word in analogy
+full_count = full_count + length(ind1);
+ind = (ind1 ~= unkkey) & (ind2 ~= unkkey) & (ind3 ~= unkkey) & (ind4 ~= unkkey); %only look at those questions which have no unknown words
+ind1 = ind1(ind);
+ind2 = ind2(ind);
+ind3 = ind3(ind);
+ind4 = ind4(ind);
+disp([filenames{j} ':']);
+mx = zeros(1,length(ind1));
+num_iter = ceil(length(ind1)/split_size);
+for jj=1:num_iter
+range = (jj-1)*split_size+1:min(jj*split_size,length(ind1));
+dist = full(W * (W(ind2(range),:)' - W(ind1(range),:)' + W(ind3(range),:)')); %cosine similarity if input W has been normalized
+for i=1:length(range)
+dist(ind1(range(i)),i) = -Inf;
+dist(ind2(range(i)),i) = -Inf;
+dist(ind3(range(i)),i) = -Inf;
+end
+[~, mx(range)] = max(dist); %predicted word index
+end
+
+val = (ind4 == mx'); %correct predictions
+count_tot = count_tot + length(ind1);
+correct_tot = correct_tot + sum(val);
+disp(['ACCURACY TOP1: ' num2str(mean(val)*100,'%-2.2f') '% (' num2str(sum(val)) '/' num2str(length(val)) ')']);
+if j < 6
+ count_sem = count_sem + length(ind1);
+ correct_sem = correct_sem + sum(val);
+else
+ count_syn = count_syn + length(ind1);
+ correct_syn = correct_syn + sum(val);
+end
+
+disp(['Total accuracy: ' num2str(100*correct_tot/count_tot,'%-2.2f') '% Semantic accuracy: ' num2str(100*correct_sem/count_sem,'%-2.2f') '% Syntactic accuracy: ' num2str(100*correct_syn/count_syn,'%-2.2f') '%']);
+
+end
+disp('________________________________________________________________________________');
+disp(['Questions seen/total: ' num2str(100*count_tot/full_count,'%-2.2f') '% (' num2str(count_tot) '/' num2str(full_count) ')']);
+disp(['Semantic Accuracy: ' num2str(100*correct_sem/count_sem,'%-2.2f') '% (' num2str(correct_sem) '/' num2str(count_sem) ')']);
+disp(['Syntactic Accuracy: ' num2str(100*correct_syn/count_syn,'%-2.2f') '% (' num2str(correct_syn) '/' num2str(count_syn) ')']);
+disp(['Total Accuracy: ' num2str(100*correct_tot/count_tot,'%-2.2f') '% (' num2str(correct_tot) '/' num2str(count_tot) ')']);
+BB = [100*correct_sem/count_sem 100*correct_syn/count_syn 100*correct_tot/count_tot];
+
+end

eval/octave/read_and_evaluate_octave.m

@@ -0,0 +1,35 @@
+addpath('./eval/octave');
+if(~exist('vocab_file')) 
+ vocab_file = 'vocab.txt';
+end
+if(~exist('vectors_file')) 
+ vectors_file = 'vectors.bin';
+end
+
+fid = fopen(vocab_file, 'r');
+words = textscan(fid, '%s %f');
+fclose(fid);
+words = words{1};
+vocab_size = length(words);
+global wordMap
+
+wordMap = struct();
+for i=1:numel(words)
+ wordMap.(words{i}) = i;
+end
+
+fid = fopen(vectors_file,'r');
+fseek(fid,0,'eof');
+vector_size = ftell(fid)/16/vocab_size - 1;
+frewind(fid);
+WW = fread(fid, [vector_size+1 2*vocab_size], 'double')'; 
+fclose(fid); 
+
+W1 = WW(1:vocab_size, 1:vector_size); % word vectors
+W2 = WW(vocab_size+1:end, 1:vector_size); % context (tilde) word vectors
+
+W = W1 + W2; %Evaluate on sum of word vectors
+W = bsxfun(@rdivide,W,sqrt(sum(W.*W,2))); %normalize vectors before evaluation
+evaluate_vectors_octave(W);
+exit
+