base

model.py

@@ -0,0 +1,38 @@
+import json
+from sentence_transformers import SentenceTransformer, util
+
+class EmojiPredictor:
+    def __init__(self, model_path, emoji_data_path):
+        self.model = SentenceTransformer(model_path)
+        self.emoji_data = self._load_emoji_data(emoji_data_path)
+        self.description_vectors = self._vectorize_descriptions()
+
+    def _load_emoji_data(self, emoji_data_path):
+        with open(emoji_data_path, 'r') as f:
+            return json.load(f)
+
+    def _vectorize_descriptions(self):
+        # Get the sentence embedding for each description
+        descriptions = [item['description'] for item in self.emoji_data]
+        return self.model.encode(descriptions)
+
+    def predict(self, text):
+        # Get the sentence embedding for the input text
+        text_vector = self.model.encode([text])[0]
+        
+        from sentence_transformers import util
+
+        # Reshape vectors for cosine similarity calculation
+        text_vector_reshaped = text_vector.reshape((1, -1))
+        
+        # Calculate cosine similarity using sentence_transformers.util.cos_sim
+        similarities = util.cos_sim(self.description_vectors, text_vector_reshaped).flatten()
+        
+        # Commented out: Manual cosine similarity calculation
+        # dot_products = np.dot(self.description_vectors, text_vector_reshaped.T).flatten()
+        # norms = np.linalg.norm(self.description_vectors, axis=1) * np.linalg.norm(text_vector_reshaped)
+        # similarities = np.divide(dot_products, norms, out=np.zeros_like(dot_products), where=norms!=0)
+        
+        # Find the index of the most similar description
+        most_similar_index = similarities.argmax()
+        return self.emoji_data[most_similar_index]['emoji']

requirements.txt

@@ -0,0 +1,2 @@
+gradio
+sentence-transformers