added search for worst cases

example_analogy/src/ofApp.cpp

@@ -111,7 +111,7 @@ void ofApp::setup(){
 
 		int count = 5;
 
-		vector<ofxWord2VecEmbeddingMatch> match = embed.match_cos(Vec, 5, used_indices);
+		vector<ofxWord2VecEmbeddingMatch> match = embed.match_cos(Vec, count, used_indices);
 		cout << "Result:" << endl;
 		for (int i = 0; i < match.size(); i++) {
 			cout << "    " << match[i].word << ": " << match[i].conf << endl;

src/ofxWord2VecEmbedding.cpp

@@ -165,7 +165,7 @@ vector<ofxWord2VecEmbeddingMatch> ofxWord2VecEmbedding::match_cos(const ofxWord2
 				for (int d = count - 1; d > a; d--) {
 					match[d] = match[d - 1];
 				}
-				match[a] = ofxWord2VecEmbeddingMatch(vocab[i], dist);
+				match[a] = ofxWord2VecEmbeddingMatch(vocab[i], dist, i);
 				break;
 			}
 		}
@@ -175,3 +175,40 @@ vector<ofxWord2VecEmbeddingMatch> ofxWord2VecEmbedding::match_cos(const ofxWord2
 }
 
 //--------------------------------------------------------------
+vector<ofxWord2VecEmbeddingMatch> ofxWord2VecEmbedding::match_worst_cos(const ofxWord2VecVector &v, int count,
+	const vector<int> &except_words) {
+
+	vector<ofxWord2VecEmbeddingMatch> worst(count);
+	for (auto &w : worst) {
+		w.conf = 2;
+	}
+
+	for (int i = 0; i < words; i++) {
+		//check if this word is allowable (not in 'except_words' array)
+		bool allow = true;
+		for (int j = 0; j < except_words.size(); j++) {
+			if (i == except_words[j]) {
+				allow = false;
+				break;
+			}
+		}
+		if (!allow) continue;
+
+		//compute distance and compare with matched
+		float dist = vec[i].dist_cosine_optimized(v);
+
+		for (int a = 0; a < count; a++) {
+			if (dist < worst[a].conf) {
+				for (int d = count - 1; d > a; d--) {
+					worst[d] = worst[d - 1];
+				}
+				worst[a] = ofxWord2VecEmbeddingMatch(vocab[i], dist, i);
+				break;
+			}
+		}
+	}
+
+	return worst;
+}
+
+//--------------------------------------------------------------

src/ofxWord2VecEmbedding.h

@@ -8,11 +8,13 @@
 //result for matching
 struct ofxWord2VecEmbeddingMatch {
 	string word;		//word name
+	int index = -1;		//word index
 	float conf = -2;	//confidence [-1..1]
 	ofxWord2VecEmbeddingMatch() {}
-	ofxWord2VecEmbeddingMatch(string word, float conf) {
+	ofxWord2VecEmbeddingMatch(string word, float conf, int index) {
 		this->word = word;
 		this->conf = conf;
+		this->index = index;
 	}
 
 };
@@ -41,10 +43,14 @@ struct ofxWord2VecEmbedding {
 	int find_case_sensitive(const string &word);
 	int find_case_insensitive(const string &word);
 
-	//find best mathing words to a given vector in cosine distance
+	//find best matching words to a given vector in cosine distance
 	vector<ofxWord2VecEmbeddingMatch> match_cos(const ofxWord2VecVector &v, int count,
 		const vector<int> &except_words = vector<int>());
 
+	//find worst matching words to a given vector in cosine distance
+	vector<ofxWord2VecEmbeddingMatch> match_worst_cos(const ofxWord2VecVector &v, int count,
+		const vector<int> &except_words = vector<int>());
+
 protected:
 	bool speedup_word_search = false;
 	map<string, int> map_case_sens;	//mapping word->index for fast word search