updating dev branch with main changes

CODEOWNERS.md

@@ -4,4 +4,5 @@ Vagner Santana - vsantana@ibm.com
 
 Cássia Sampaio - csamp@ibm.com
 
-Ashwath Vaithinathan Aravindan - ashwath.vaithina@ibm.com
+Ashwath Vaithinathan Aravindan - ashwath.vaithina@ibm.com
+

SECURITY.md

@@ -7,11 +7,12 @@ currently being supported with security updates.
 
 | Version | Supported          |
 | ------- | ------------------ |
-| 5.1.x   | :white_check_mark: |
-| 5.0.x   | :x:                |
+| v0.1.0  | :white_check_mark: |
+<!--| 5.0.x   | :x:                |
 | 4.0.x   | :white_check_mark: |
-| < 4.0   | :x:                |
+| < 4.0   | :x:                |-->
 
 ## Reporting a Vulnerability
 
-To report a security issue, please email $VMTalias with a description of the issue, the steps you took to create the issue, affected versions, and if known, mitigations for the issue. Our vulnerability management team will acknowledge receiving your email within 3 working days. This project follows a 90 day disclosure timeline.
+To report a security issue, please email us with a description of the issue, the steps you took to create the issue, affected versions, and if known, mitigations for the issue. Our vulnerability management team will acknowledge receiving your email within 3 working days. This project follows a 90 day disclosure timeline.
+

app.py

@@ -38,6 +38,8 @@
 import json
 import os
 import pickle
+import numpy as np
+from functools import lru_cache
 
 app = Flask(__name__)
 
@@ -57,6 +59,41 @@
 app.register_blueprint(cfg.SWAGGER_BLUEPRINT, url_prefix = cfg.SWAGGER_URL)
 FRONT_LOG_FILE = 'front_log.json'
 
+@lru_cache(maxsize=1)
+def get_values_embedding_function():
+    """
+    Getting the embedding function for the /values endpoint.
+    Cached to avoid reloading the model multiple times.
+
+    Returns:
+        Embedding function callable
+    """
+    model_id, model_path = save_model.save_model()
+    return recommendation_handler.get_embedding_func(inference='local', model_id=model_path)
+
+@lru_cache(maxsize=1)
+def get_values_centroids():
+    """
+    Getting the positive and negative value centroids for the /values endpoint.
+    Cached to avoid reloading the data multiple times.
+
+    Returns:
+        Dictionary with 'positive' and 'negative' centroid embeddings
+    """
+    prompt_json = recommendation_handler.populate_json()
+    positive_category_centroid = {}
+    negative_category_centroid = {}
+
+    for category in prompt_json['positive_values']:
+        positive_category_centroid[category['label']] = np.array(category['centroid'])
+
+    for category in prompt_json['negative_values']:
+        negative_category_centroid[category['label']] = np.array(category['centroid'])
+
+    return {
+        'positive': positive_category_centroid,
+        'negative': negative_category_centroid
+    }
 
 @app.route("/")
 def index():
@@ -109,7 +146,7 @@ def get_thresholds():
 @cross_origin()
 def recommend_local():
     model_id, _ = save_model.save_model()
-    prompt_json, _ = recommendation_handler.populate_json()
+    prompt_json = recommendation_handler.populate_json()
     args = request.args
     print("args list = ", args)
     prompt = args.get("prompt")
@@ -166,6 +203,42 @@ def demo_inference():
         return response
     except:
         return "Model Inference failed.", 500
+
+@app.route("/values", methods=['GET'])
+@cross_origin()
+def get_values():
+    """
+    Getting positive and negative values for a given prompt using cached embedding function and centroids for performance.
+    """
+    args = request.args
+    prompt = args.get("prompt")
+
+    # validating input
+    if not prompt:
+        return jsonify({"error": "Missing required parameter: prompt"}), 400
+
+    if not isinstance(prompt, str):
+        return jsonify({"error": "Parameter 'prompt' must be a string"}), 400
+
+    if len(prompt.strip()) == 0:
+        return jsonify({"error": "Parameter 'prompt' cannot be empty"}), 400
+
+    try:
+        embedding_fn = get_values_embedding_function()
+        centroids = get_values_centroids()
+
+        values = recommendation_handler.get_values(
+            prompt, 
+            centroids['positive'], 
+            centroids['negative'],
+            embedding_fn
+        )
+
+        return jsonify(values)
+
+    except Exception as e:
+        logger.error(f'Error in /values endpoint: {str(e)}')
+        return jsonify({"error": "Internal server error processing prompt"}), 500
 
 if __name__=='__main__':
     debug_mode = os.getenv('FLASK_DEBUG', 'False').lower() in ['true', '1', 't']

control/recommendation_handler.py

@@ -335,3 +335,82 @@ def get_thresholds(
     thresholds['remove_higher_threshold'] = round(remove_similarities_df.describe([.9]).loc['90%', 'similarity'], 1)
 
     return thresholds
+
+def get_values(
+    prompt,
+    positive_embeddings,
+    negative_embeddings,
+    embedding_fn = None
+):
+    """
+    Compute positive and negative value associations for each sentence in the input prompt.
+
+    Args:
+        prompt: Input prompt text.
+        positive_embeddings: Dictionary mapping positive value labels to centroid embeddings.
+        negative_embeddings: Dictionary mapping negative value labels to centroid embeddings.
+        embedding_fn: Function to generate embeddings from text.
+
+    Returns:
+        Dictionary containing sentences with their associated positive and negative values and similarity scores.
+    """
+
+    if embedding_fn is None:
+        # using all-MiniLM-L6-v2 locally by default
+        embedding_fn = get_embedding_func('local', model_id='sentence-transformers/all-MiniLM-L6-v2')
+
+    sentences = split_into_sentences(prompt)
+
+    # bifurcating and filtering out empty sentences
+    sentences = [s for s in sentences if s.strip()]
+
+    values = {}
+    values["prompts"] = []
+
+    # returning if no valid sentences
+    if not sentences:
+        return values
+
+    # generating all sentence embeddings in a single call by batching all
+    sentence_embeddings = embedding_fn(sentences)
+    sentence_embeddings = np.array(sentence_embeddings)
+
+    # ensuring embeddings have correct shape - expanding embeddings of all sentences
+    if len(sentence_embeddings.shape) == 1:
+        sentence_embeddings = np.expand_dims(sentence_embeddings, axis=0)
+
+    # processing each sentence with its corresponding embedding
+    for idx, sentence in enumerate(sentences):
+
+        sentence_embedding = sentence_embeddings[idx]
+
+        max_similarity_positive = -1
+        positive_label = None
+        for label, centroid in positive_embeddings.items():
+            similarity = cosine_similarity(
+                np.expand_dims(sentence_embedding, axis=0),
+                np.array([centroid])
+            )[0, 0]
+            if similarity > max_similarity_positive:
+                max_similarity_positive = similarity
+                positive_label = label
+
+        max_similarity_negative = -1
+        negative_label = None
+        for label, centroid in negative_embeddings.items():
+            similarity = cosine_similarity(
+                np.expand_dims(sentence_embedding, axis=0),
+                np.array([centroid])
+            )[0, 0]
+            if similarity > max_similarity_negative:
+                max_similarity_negative = similarity
+                negative_label = label
+
+        values["prompts"].append({
+            "sentence": sentence,
+            "positive_value": {"label": positive_label, "similarity": float(max_similarity_positive)},
+            "negative_value": {"label": negative_label, "similarity": float(max_similarity_negative)}
+        })
+
+    return values
+