add config methods to bert connective disambiguation model

Author: rknaebel

discopy/components/connective/bert.py

@@ -70,18 +70,33 @@ class ConnectiveClassifier(Component):
     def __init__(self, input_dim, used_context: int = 0):
         in_size = input_dim + 2 * used_context * input_dim
         self.model = get_conn_model(in_size, 1, 1024)
+        self.input_dim = input_dim
         self.used_context = used_context
 
+    def get_config(self):
+        return {
+            'model_name': self.model_name,
+            'input_dim': self.input_dim,
+            'used_context': self.used_context,
+        }
+
+    @staticmethod
+    def from_config(config: dict):
+        clf = ConnectiveClassifier(config['input_dim'], config['used_context'])
+        clf.sense_map = config['sense_map']
+        clf.classes = config['classes']
+        return clf
+
     def load(self, path):
-        if not os.path.exists(os.path.join(path, f'connective_nn_{self.used_context}.model')):
+        if not os.path.exists(os.path.join(path, self.model_name)):
             raise FileNotFoundError("Model not found.")
-        self.model = tf.keras.models.load_model(os.path.join(path, f'connective_nn_{self.used_context}.model'),
+        self.model = tf.keras.models.load_model(os.path.join(path, self.model_name),
                                                 compile=False)
 
     def save(self, path):
         if not os.path.exists(path):
             os.makedirs(path)
-        self.model.save(os.path.join(path, f'connective_nn_{self.used_context}.model'))
+        self.model.save(os.path.join(path, self.model_name))
 
     def fit(self, docs_train: List[Document], docs_val: List[Document] = None):
         if docs_val is None: