DEPRESSION-DETECTION

source_code/app_utilities.py

@@ -38,28 +38,50 @@ def tweet_prediction(tweet: str) -> int:
     Returns:
         int: 1 for Depressive, 0 for Non-depressive.
     """
-    # Step 1: Clean the text
-    processed_tweet = tweet
-    cleaned_input = []
-    cleaned_input.append(CU.tweets_cleaner(processed_tweet))
-    
-    # Step 2: Convert text to numbers using spaCy
+# Global initialization of heavy resources to optimize runtime performance
+# Loading these once at startup eliminates significant latency during individual requests
+
+# 1. Load spaCy NLP engine
+try:
     nlp_engine = en_core_web_lg.load()
-    
-    # Step 3: Compute centroid word embeddings
-    # We calculate the mean vector of all tokens to represent the tweet's semantic context
-    semantic_vectors = np.array([
-        np.array([token.vector for token in nlp_engine(s)]).mean(axis=0) * np.ones((300))
-        for s in cleaned_input
-    ])
-    
-    # Step 4: Load the pre-trained Support Vector Machine (SVM) model artifact
-    # The SVM was selected for its robust performance in high-dimensional text classification
-    model_path = "./assets/models/model_svm1.pkl"
+except Exception as e:
+    print(f"Error loading spaCy model: {e}")
+    sys.exit(1)
+
+# 2. Load pre-trained SVM Classifier
+model_path = "./assets/models/model_svm1.pkl"
+try:
     with open(model_path, 'rb') as model_file:
         classifier = pickle.load(model_file)
+except Exception as e:
+    print(f"Error loading SVM model from {model_path}: {e}")
+    sys.exit(1)
+
+def tweet_prediction(tweet: str) -> int:
+    """
+    Takes a tweet and returns whether it's classified as depressive (1) or not (0).
+    
+    The process:
+        1. Clean the text using our utility module.
+        2. Convert text to numbers using the pre-loaded spaCy engine.
+        3. Use the pre-loaded SVM model to make a prediction.
+    Args:
+        tweet (str): The tweet text from the user.
+
+    Returns:
+        int: 1 for Depressive, 0 for Non-depressive.
+    """
+    # Step 1: Clean the text
+    cleaned_text = CU.tweets_cleaner(tweet)
+    
+    # Step 2: Compute centroid word embeddings
+    # We calculate the mean vector of all tokens to represent the tweet's semantic context
+    # Note: Global 'nlp_engine' is used here, avoiding reload overhead
+    vector = np.array([token.vector for token in nlp_engine(cleaned_text)]).mean(axis=0) * np.ones((300))
+    semantic_vectors = np.array([vector])
     
-    # Step 5: Perform binary classification
+    # Step 3: Perform binary classification
+    # Note: Global 'classifier' is used here
     prediction_result = classifier.predict(semantic_vectors)
     
     return int(prediction_result[0])

source_code/requirements.txt

@@ -35,3 +35,4 @@ Flask-Bootstrap==3.3.7.1
 Jinja2==3.0.3
 Werkzeug==2.0.3
 itsdangerous==2.0.1
+