classifier: fix memory usage

inspire_classifier/api.py

@@ -151,24 +151,30 @@ def train():
     train_and_save_classifier()
 
 
-def predict_coreness(title, abstract):
+def initialize_classifier():
     """
-    Predicts class-wise probabilities given the title and abstract.
+    Initializes the classifier.
     """
-    text = title + " <ENDTITLE> " + abstract
-    categories = ["rejected", "non_core", "core"]
-    try:
-        classifier = Classifier(
-            cuda_device_id=current_app.config["CLASSIFIER_CUDA_DEVICE_ID"]
-        )
-    except IOError as error:
-        raise IOError("Data ITOS not found.") from error
-
+    classifier = Classifier(
+        cuda_device_id=current_app.config["CLASSIFIER_CUDA_DEVICE_ID"]
+    )
     try:
         classifier.load_trained_classifier_weights(path_for("trained_classifier"))
     except IOError as error:
-        raise IOError("Could not load the trained classifier weights.") from error
+        raise IOError(
+            "Could not load the trained classifier weights.",
+            path_for("trained_classifier"),
+        ) from error
 
+    return classifier
+
+
+def predict_coreness(classifier, title, abstract):
+    """
+    Predicts class-wise probabilities given the title and abstract.
+    """
+    text = title + " <ENDTITLE> " + abstract
+    categories = ["rejected", "non_core", "core"]
     class_probabilities = classifier.predict(
         text, temperature=current_app.config["CLASSIFIER_SOFTMAX_TEMPERATUR"]
     )
@@ -191,17 +197,22 @@ def validate(validation_df):
         raise IOError("There was a problem loading the classifier model") from error
     predictions = []
     validation_df = validation_df.sample(frac=1, random_state=42)
-    for _, row in tqdm(
-        validation_df.iterrows(), total=len(validation_df.label.values)
-    ):
+    for _, row in tqdm(validation_df.iterrows(), total=len(validation_df.label.values)):
         predicted_value = classifier.predict(
             row.text, temperature=current_app.config["CLASSIFIER_SOFTMAX_TEMPERATUR"]
         )
         predicted_class = np.argmax(predicted_value)
         predictions.append(predicted_class)
 
-    validation_df.insert(2, 'predicted_label', predictions)
+    validation_df.insert(2, "predicted_label", predictions)
     validation_df.to_csv(f"{path_for('data')}/validation_results.csv", index=False)
-    print("f1 score ", f1_score(validation_df["label"], validation_df["predicted_label"], average="micro"))
-    pprint(classification_report(validation_df["label"], validation_df["predicted_label"]))
+    print(
+        "f1 score ",
+        f1_score(
+            validation_df["label"], validation_df["predicted_label"], average="micro"
+        ),
+    )
+    pprint(
+        classification_report(validation_df["label"], validation_df["predicted_label"])
+    )
     pprint(confusion_matrix(validation_df["label"], validation_df["predicted_label"]))

inspire_classifier/app.py

@@ -28,7 +28,7 @@
 from prometheus_flask_exporter.multiprocess import GunicornInternalPrometheusMetrics
 from webargs.flaskparser import use_args
 
-from inspire_classifier.api import predict_coreness
+from inspire_classifier.api import initialize_classifier, predict_coreness
 
 from . import serializers
 
@@ -55,6 +55,8 @@ def create_app():
     app.config["CLASSIFIER_BASE_PATH"] = app.instance_path
     app.config.from_object("inspire_classifier.config")
     app.config.from_pyfile("classifier.cfg", silent=True)
+    with app.app_context():
+        classifier = initialize_classifier()
 
     @app.route("/api/health")
     def date():
@@ -69,7 +71,7 @@ def date():
     )
     def core_classifier(args):
         """Endpoint for the CORE classifier."""
-        prediction = predict_coreness(args["title"], args["abstract"])
+        prediction = predict_coreness(classifier, args["title"], args["abstract"])
         response = coreness_schema.dump(prediction)
         return response
 

scripts/create_dataset.py

@@ -92,8 +92,13 @@ def __init__(self, index, query_filters, year_from, year_to, month_from, month_t
             self.inspire_categories_field = "inspire_categories.term"
             self.query_filters = [
                 query_filters
-                & Q("range", _created={"gte": f"{self.year_from}-{self.month_from}",
-                                       "lt": f"{self.year_to}-{self.month_to}",}),
+                & Q(
+                    "range",
+                    _created={
+                        "gte": f"{self.year_from}-{self.month_from}",
+                        "lt": f"{self.year_to}-{self.month_to}",
+                    },
+                ),
             ]
 
     def _postprocess_record_data(self, record_data):
@@ -143,7 +148,9 @@ def prepare_inspire_classifier_dataset(data, save_data_path):
     inspire_data_df["text"] = (
         inspire_data_df["title"] + " <ENDTITLE> " + inspire_data_df["abstract"]
     )
-    inspire_classifier_data_df = inspire_data_df[["id", "inspire_categories", "label", "text"]]
+    inspire_classifier_data_df = inspire_data_df[
+        ["id", "inspire_categories", "label", "text"]
+    ]
     inspire_classifier_data_df.to_pickle(save_data_path)
 
 
@@ -163,7 +170,8 @@ def get_inspire_classifier_dataset(year_from, year_to, month_from, month_to):
     month_to = f"{month_to:02d}-31"
     print(f"Fetching {year_from}-{month_from} to {year_to}-{month_to}")
     inspire_classifier_dataset_path = os.path.join(
-        os.getcwd(), f"inspire_classifier_dataset_{year_from}-{month_from}_{year_to}-{month_to}.pkl"
+        os.getcwd(),
+        f"inspire_classifier_dataset_{year_from}-{month_from}_{year_to}-{month_to}.pkl",
     )
     data = get_data_for_decisions(year_from, year_to, month_from, month_to)
     prepare_inspire_classifier_dataset(data, inspire_classifier_dataset_path)