Merge pull request #8 from Boeing/hai-branch

Added Fire, PDA and RTO models

src/sdr_hazards_classification/prep_utils.py

@@ -15,6 +15,9 @@
 DEPRESSURIZATION = 'depressurization'
 DEGRADED_CONTROLLABILITY= 'degraded-controllability'
 CORROSION_LIMIT = 'corrosion-limit'
+FIRE = 'fire'
+PDA = 'pda'
+RTO = 'rto'
 
 LOG_FILE = os.getcwd() + "/logs"
 if not os.path.exists(LOG_FILE):

src/sdr_hazards_classification/sdr_api.py

@@ -10,7 +10,7 @@
 
 import numpy as np
 
-from .prep_utils import PreprocessingUtils, DEPRESSURIZATION, DEGRADED_CONTROLLABILITY, CORROSION_LIMIT
+from .prep_utils import PreprocessingUtils, DEPRESSURIZATION, DEGRADED_CONTROLLABILITY, CORROSION_LIMIT, FIRE, PDA, RTO
 from .vectorizers import Vectorizers
 import pandas as pd
 
@@ -27,6 +27,15 @@ def __init__(self, event_type="depressurization"):
         elif event_type in DEPRESSURIZATION:
             model_type = f"sdr-{DEPRESSURIZATION}.model"
             model_config = f"sdr-{DEPRESSURIZATION}.config"
+        elif event_type in FIRE:
+            model_type = f"sdr-{FIRE}.model"
+            model_config = f"sdr-{FIRE}.config"
+        elif event_type in RTO:
+            model_type = f"sdr-{RTO}.model"
+            model_config = f"sdr-{RTO}.config"
+        elif event_type in PDA:
+            model_type = f"sdr-{PDA}.model"
+            model_config = f"sdr-{PDA}.config"
         else: assert "Event type not supported!"
 
         this_dir, this_filename = os.path.split(__file__)  # Get path of data.pkl
@@ -121,8 +130,11 @@ def test_sdr_corrosion_limit(self):
 if __name__ == "__main__":
 
    #Load the trained model
-   model_api = SdrInferenceAPI(event_type='degraded-controllability')
-   model_api.test_sdr_degraded_controllability()
+   model_api = SdrInferenceAPI(event_type=CORROSION_LIMIT)
+   # model_api.test_sdr_degraded_controllability()
    model_api.test_sdr_corrosion_limit()
-   model_api.test_sdr_depressurization_predictions()
+   # model_api.test_sdr_depressurization_predictions()
 
+   fire_model = SdrInferenceAPI(event_type=PDA)
+   text = """strong odor of smoke in the flight deck. troubleshooting showed main deck cargo pdu 9r motor burned, tripping the circuit-breaker. and 10r overheated. both pdu ' s removed and cover plate installed"""
+   print(fire_model.get_predictions([text]))

src/sdr_hazards_classification/sdr_classifier.py

@@ -15,10 +15,10 @@
 from os import path
 import pickle
 import traceback as tb
-# Default column name  in DataFrame
+from sklearn.model_selection import cross_val_score
 from sklearn.svm import SVC
 
-from prep_utils import PreprocessingUtils, DEPRESSURIZATION, DEGRADED_CONTROLLABILITY, CORROSION_LIMIT
+from prep_utils import PreprocessingUtils, DEPRESSURIZATION, DEGRADED_CONTROLLABILITY, CORROSION_LIMIT, FIRE, RTO, PDA
 from vectorizers import Vectorizers
 
 prep_util = PreprocessingUtils()
@@ -106,30 +106,34 @@ def train_text_classifier(criteria, all_df , text_field, label_field, model_type
         X_train = Vectorizers.transform_with_vectorizers(train_data, vectorizers)
         model_config["vectorizers"] = vectorizers
 
-        model = None
+        sklearn_model = None
         if model_config["model_name"] == "lg":
-            model = LogisticRegression(solver='liblinear', C=1e2, max_iter=200)
+            sklearn_model = LogisticRegression(solver='liblinear', C=1e2, max_iter=200)
         elif model_config["model_name"] == "svm":
-            model = SVC(kernel='linear', C=1.0)
+            sklearn_model = SVC(kernel='linear', C=1.0)
         elif model_config["model_name"] == "xgb":
             from xgboost import XGBClassifier
-            model = XGBClassifier(n_estimators=500, max_depth=5, reg_alpha=.1)
+            sklearn_model = XGBClassifier(n_estimators=1000, max_depth=10, reg_alpha=.1)
         else:
             model_name = model_config["model_name"]
             raise ValueError(f"Model '{model_name}' not supported")
 
         # fit model
-        model.fit(X_train, train_labels)
+        print(f"Let's do CV for {type(sklearn_model).__name__}")
+        scores = cross_val_score(sklearn_model, X_train, train_labels, scoring= 'accuracy', cv=5)
+        print("%0.2f accuracy with a standard deviation of %0.2f" % (scores.mean(), scores.std()))
+
+        sklearn_model.fit(X_train, train_labels)
 
         if evaluate_model:
             # target_names = [criteria, 'Non-'+criteria]
-            target_name = le.inverse_transform(model.classes_)
-            predictions = TextClassifier.predict_sdr_labels(model, model_config, test_data)
+            target_name = le.inverse_transform(sklearn_model.classes_)
+            predictions = TextClassifier.predict_sdr_labels(sklearn_model, model_config, test_data)
             print(classification_report(test_labels, predictions[0], target_names=target_name))
             logging.info(msg=f'Performance:\n {classification_report(test_labels, predictions[0], target_names=target_name)}')
-            return model, model_config, test_data, test_labels
+            return sklearn_model, model_config, test_data, test_labels
 
-        return model, model_config
+        return sklearn_model, model_config
 
     @staticmethod
     def export_model(model, model_config, model_name, out_folder):
@@ -233,18 +237,33 @@ def release_model( **kwargs):
 
     if args.event_type in DEPRESSURIZATION:
         training_args = {'event_type': DEPRESSURIZATION,
-                    'model_type': 'lg',
+                    'model_type': args.model_type,
                    'target_name': ["depressurization", "non-depressurization"],
                    'training_path': "./data/Depressurization.csv"}
+    elif args.event_type in FIRE:
+        training_args = {'event_type': FIRE,
+                    'model_type': args.model_type,
+                   'target_name': ["fire", "non-fire"],
+                   'training_path': "./data/Fire.csv"}
+    elif args.event_type in PDA:
+        training_args = {'event_type': PDA,
+                    'model_type': args.model_type,
+                   'target_name': ["pda", "non-pda"],
+                   'training_path': "./data/Parts_Departing_Aircraft.csv"}
+    elif args.event_type in RTO:
+        training_args = {'event_type': RTO,
+                    'model_type': args.model_type,
+                   'target_name': ["rto", "non-rto"],
+                   'training_path': "./data/Reject_To_Takeoff.csv"}
     elif args.event_type in DEGRADED_CONTROLLABILITY:
         training_args = {'event_type': DEGRADED_CONTROLLABILITY,
-                    'model_type': 'xgb',
+                    'model_type': args.model_type,
                    'target_name': ["degraded-controllability", "non-degraded"],
                    'training_path': "./data/Degraded_Controllability.csv"}
     elif args.event_type in CORROSION_LIMIT:
         training_args = {
             'event_type': CORROSION_LIMIT,
-            'model_type': 'lg',
+            'model_type': args.model_type,
             'target_name': ["beyond-limit", "no-corrosion", "no-limit", "within-limit", "within-beyond-limit"],
             'training_path': "./data/Corrosion_Limit.csv"}
     else:

src/test/test_model.py

@@ -28,6 +28,18 @@ def load_controll_model(self, event_type="degraded-controllability"):
     def load_corrosion_model(self, event_type="corrosion-limit"):
         yield SdrInferenceAPI(event_type)
 
+    @pytest.fixture
+    def load_fire_model(self, event_type="fire"):
+        yield SdrInferenceAPI(event_type)
+
+    @pytest.fixture
+    def load_pda_model(self, event_type="pda"):
+        yield SdrInferenceAPI(event_type)
+
+    @pytest.fixture
+    def load_rto_model(self, event_type="rto"):
+        yield SdrInferenceAPI(event_type)
+
     def test_depressurization_prediction_No(self, load_depressurization_model):
 
         expect_result = np.array(["No", "No"])
@@ -97,6 +109,51 @@ def test_controllability_prediction_Yes(self, load_controll_model):
         assert_array_equal(pred, expect_result)
         assert np.all(np.array(probs) > 0.9)
 
+    def test_fire(self, load_fire_model):
+
+        expect_result = np.array(["No", "Yes", "No", "Yes"])
+
+        record = ["""RETURNED TO DEPARTURE DUE TO THE LEFT ENGINE OIL QUANTITY WAS DECREASING. REPLACED THE ENGINE. 
+            Nature of Condition:  WARNING INDICATION, FLUID LOSS Precautionary Procedure:  UNSCHED LANDING Part Name: OIL SYSTEM Part Condition: LEAKING""",
+                  """burning electrical odor noticed in cabin from row 28 to back of aircraft. no signs of smoke noticed. odor approximately 1 hr out from landing. ran nr 1 and 2 engines independently with packs running. no smoke or any burning odor. checked reading lights iaw mcc for shorting and arcing. no signs of shorting and arcing. checked ovens and coffeemakers, no evidence of burning or arcing noted. nature of condition : smoke fumes odors sparks precautionary procedure : other part name : unknown part condition : odor.""",
+                  """CABIN ALTITUDE EICAS WARNING AT FL380. COMPLIED WITH QRH,  DESCENDED TO FINAL ALTITUDE FL250. RECOVERED CABIN PRESSURE IN AUTO 2 AT CABIN ALTITUDE 7500.  NORMAL DESCENT AND PRESSURE UNTIL LFPG.   REPLACED THE NR 1 CPC IAW MM 21-31-02-4, TESTED OK.  COMPLYED WITH CABIN PRESSURE DECAY CHECK IAW MM 05-51-24-2, TEST PASSED. Nature of Condition:  
+                WARNING INDICATION Precautionary Procedure:  O2 MASK DEPLOYED Part Name: CONTROLLER Part Condition: MALFUNCTIONED""",
+                  """on climbout the cockpit filled with smoke. the aircraft declared an emergency and returned to the airport. the smoke was observed coming from first officer ' s side. after aircraft returned to gate, mechanic found p6-12 circuit-breaker with evidence of being overheated and terminal connector burned. nature of condition : smoke fumes odors sparks precautionary procedure : unsched landing part name : circuit-breaker part condition : failed"""
+                  ]
+
+        pred, probs = load_fire_model.get_predictions(record)
+
+        assert_array_equal(pred, expect_result)
+        assert np.all(np.array(probs) > 0.9)
+
+    def test_pda(self, load_pda_model):
+        expect_result = np.array(["No", "Yes", "No", "Yes"])
+
+        record = ["""CABIN ALTITUDE EICAS WARNING AT FL380. COMPLIED WITH QRH,  DESCENDED TO FINAL ALTITUDE FL250. RECOVERED CABIN PRESSURE IN AUTO 2 AT CABIN ALTITUDE 7500.  NORMAL DESCENT AND PRESSURE UNTIL LFPG.   REPLACED THE NR 1 CPC IAW MM 21-31-02-4, TESTED OK.  COMPLYED WITH CABIN PRESSURE DECAY CHECK IAW MM 05-51-24-2, TEST PASSED. Nature of Condition:  
+        WARNING INDICATION Precautionary Procedure:  O2 MASK DEPLOYED Part Name: CONTROLLER Part Condition: MALFUNCTIONED""",
+        """right engine pylon aft outboard side panel assembly is missing panel number 446al. the panel possibly departed during op eration of the aircraft. the airplane was removed from service ( grounded ) and a serviceable panel 446al was installed. the cause of the missing panel is unknown.""",
+                  """aircraft was not grounded : during climb, autopilot engaged for approximately one second then began to roll left, 
+                  followed by autopilot disconnect and aural warning on two attempts. replaced autopilot aileron switch iaw amm 22-11-12, operation check good.""",
+                  """( log 5515918 parts departing airplane ) during pre-flight discovered large panel missing forward of right main wheel well. maintenance reinstalled body panel 192br per amm 53-51."""]
+
+        pred, probs = load_pda_model.get_predictions(record)
+
+        assert_array_equal(pred, expect_result)
+        assert np.all(np.array(probs) > 0.9)
+
+    def test_rto(self, load_rto_model):
+        expect_result = np.array(["No", "Yes", "No", "Yes"])
+
+        record = ["""forward cargo bay, bs 478 - 500, stringer 26l found cracked. removed the damaged stringer section and fabricated a repair iaw srm 53-00-03, fig 201, repair 2. installed iaw srm 51-40-02. nature of condition : other precautionary procedure : none part name : stringer part condition : cracked""",
+        """no airspeed on captains indicator, aborted takeoff and returned to blocks. inspected pitot tube and removed debris, op ' s and leak check good iaw amm 34-11-00. nature of condition : other precautionary procedure : aborted takeoff part name : pitot tube part condition : obstructed""",
+                  """during inspection, found a crack on the floor support between bs 277 and 294.5, rbl 1. maintenance replaced the floor support iaw srm 51-40-02.""",
+                  """during takeoff roll, the left thrust reverser warning light illuminated. aborted takeoff at approximately 40 knots. removed and replaced the engine accessory unit iaw amm 75-34-06. accomplished built in test equipment check iaw fim 78-31-00. system checked normal. nature of condition : warning indication precautionary procedure : aborted takeoff part name : accessory unit part condition : faulty"""]
+
+        pred, probs = load_rto_model.get_predictions(record)
+
+        assert_array_equal(pred, expect_result)
+        assert np.all(np.array(probs) > 0.9)
+
     def test_corrosion_limit(self, load_corrosion_model):
         expect_result = np.array(["Beyond Limit", "No Corrosion", "No Limit", "Within Limit", "Within_Beyond_Limit"])