Save models in seprate directories

Instead of grouping PE and AP models in the predicate_extraction and
argument_prediction directories, models now now have a individual
directory, which will contain those two subdirectories inside.

triple_extractor_ptbr_pligabue/argument_prediction/model.py

@@ -3,7 +3,7 @@
 
 from typing import cast, Optional
 
-from ..constants import DEFAULT_SENTENCE_SIZE, ARGUMENT_PREDICTION_MODEL_DIR, DEFAULT_MODEL_NAME
+from ..constants import DEFAULT_SENTENCE_SIZE, MODEL_DIR, ARGUMENT_PREDICTION_MODEL_DIR_NAME, DEFAULT_MODEL_NAME
 from ..bert import bert
 from ..predicate_extraction.types import ArgPredInputs
 
@@ -21,12 +21,20 @@ def __init__(self, *layer_units: int, name: Optional[str] = None,
             super().__init__(sentence_size)
             self._config_model(*layer_units)
 
+    @staticmethod
+    def full_model_path(name: str):
+        return MODEL_DIR / name
+
+    @staticmethod
+    def ap_model_path(name: str):
+        return ArgumentPredictor.full_model_path(name) / ARGUMENT_PREDICTION_MODEL_DIR_NAME
+
     @classmethod
     def load(cls, name: str = DEFAULT_MODEL_NAME):
         return cls(name=name)
 
     def _load_model(self, name: str):
-        path = ARGUMENT_PREDICTION_MODEL_DIR / name
+        path = self.ap_model_path(name)
         if path.is_dir():
             model = tf.keras.models.load_model(path)
             self.model = cast(tf.keras.Model, model)
@@ -70,9 +78,9 @@ def _config_model(self, *layer_units: int):
         self.model.layers[2].trainable = False
 
     def save(self, name: str):
-        path = ARGUMENT_PREDICTION_MODEL_DIR / name
-        if not ARGUMENT_PREDICTION_MODEL_DIR.is_dir():
-            ARGUMENT_PREDICTION_MODEL_DIR.mkdir()
+        self.full_model_path(name).mkdir(exist_ok=True)
+        path = self.ap_model_path(name)
+        path.mkdir(exist_ok=True)
         self.model.save(path)
 
     def compile(self, optimizer=None, loss=None, metrics=None):

triple_extractor_ptbr_pligabue/constants.py

@@ -7,8 +7,8 @@
 OBJECT_PATTERN = r"<OBJECT>(.*)</OBJECT>"
 SPECIAL_TOKEN_IDS = [101, 102, 0]
 MODEL_DIR = Path(__file__).parent / "models"
-PREDICATE_EXTRACTION_MODEL_DIR = MODEL_DIR / "predicate_extraction"
-ARGUMENT_PREDICTION_MODEL_DIR = MODEL_DIR / "argument_prediction"
+PREDICATE_EXTRACTION_MODEL_DIR_NAME = "predicate_extraction"
+ARGUMENT_PREDICTION_MODEL_DIR_NAME = "argument_prediction"
 DEFAULT_MODEL_NAME = "default"
 DEFAULT_PRED_THRESHOLD = 0.2
 DEFAULT_ARG_THREHSOLD = 0.15

triple_extractor_ptbr_pligabue/predicate_extraction/model.py

@@ -3,7 +3,7 @@
 
 from typing import cast, Optional
 
-from ..constants import DEFAULT_SENTENCE_SIZE, PREDICATE_EXTRACTION_MODEL_DIR, DEFAULT_MODEL_NAME
+from ..constants import DEFAULT_SENTENCE_SIZE, MODEL_DIR, PREDICATE_EXTRACTION_MODEL_DIR_NAME, DEFAULT_MODEL_NAME
 from ..bert import bert
 from .constants import ACCEPTANCE_THRESHOLD, O_THRESHOLD
 from .data_formatter import DataFormatter
@@ -20,16 +20,23 @@ def __init__(self, *dense_layer_units: int, name: Optional[str] = None,
             super().__init__(sentence_size)
             self._config_model(*dense_layer_units)
 
+    @staticmethod
+    def full_model_path(name: str):
+        return MODEL_DIR / name
+
+    @staticmethod
+    def pe_model_path(name: str):
+        return PredicateExtractor.full_model_path(name) / PREDICATE_EXTRACTION_MODEL_DIR_NAME
+
     @classmethod
     def load(cls, name: str = DEFAULT_MODEL_NAME):
-        path = PREDICATE_EXTRACTION_MODEL_DIR / name
-        if path.is_dir():
+        if cls.pe_model_path(name).is_dir():
             return cls(name=name)
         else:
             raise Exception(f"Model {str} does not exist.")
 
     def _load_model(self, name: str):
-        path = PREDICATE_EXTRACTION_MODEL_DIR / name
+        path = self.pe_model_path(name)
         if path.is_dir():
             model = tf.keras.models.load_model(path)
             self.model = cast(tf.keras.Model, model)
@@ -52,9 +59,9 @@ def _config_model(self, *dense_layer_units: int):
         self.model.layers[1].trainable = False
 
     def save(self, name: str = DEFAULT_MODEL_NAME):
-        path = PREDICATE_EXTRACTION_MODEL_DIR / name
-        if not PREDICATE_EXTRACTION_MODEL_DIR.is_dir():
-            PREDICATE_EXTRACTION_MODEL_DIR.mkdir()
+        self.full_model_path(name).mkdir(exist_ok=True)
+        path = self.pe_model_path(name)
+        path.mkdir(exist_ok=True)
         self.model.save(path)
 
     def compile(self, optimizer=None, loss=None, metrics=None):