Upload 4 files

app.py

@@ -70,6 +70,7 @@ st.sidebar.title("Recent Questions")
 recent_data = load_recent_questions()
 for q in reversed(recent_data["questions"]):  # Show latest first
     st.sidebar.write(f"🔹 {q['question']}")
+    st.json({q['metrics']})
 
 st.sidebar.markdown("---")  # Separator
 
@@ -80,15 +81,15 @@ import matplotlib.pyplot as plt
 # for visualization
 st.sidebar.title("Analytics")
 
-context_relevance = [q["metrics"]["context_relevance"] for q in recent_data["questions"]]
+#context_relevance = [q["metrics"]["context_relevance"] for q in recent_data["questions"]]
 response_time = [q["metrics"]["response_time"] for q in recent_data["questions"]]
-labels = [f"Q{i+1}" for i in range(len(context_relevance))]  # Labels for X-axis
+labels = [f"Q{i+1}" for i in range(len(response_time))]  # Labels for X-axis
 
 fig, ax = plt.subplots()
-ax.plot(labels, context_relevance, marker="o", label="Context Relevance")
-ax.plot(labels, response_time, marker="s", label="Response Time (sec)")
+#ax.plot(labels, context_relevance, marker="o", label="Context Relevance")
+#ax.plot(labels, response_time, marker="s", label="Response Time (sec)")
 ax.set_xlabel("Recent Questions")
-ax.set_ylabel("Scores")
+ax.set_ylabel("Time Taken for Response")
 ax.legend()
 st.sidebar.pyplot(fig)
 

data_processing.py

@@ -33,6 +33,7 @@ ragbench = {}
 index = None  
 chunk_docs = []
 documents = [] 
+query_dataset_data = {}
 
 # Ensure data directory exists
 os.makedirs("data_local", exist_ok=True)
@@ -92,6 +93,14 @@ def load_ragbench():
         ragbench[dataset] = load_dataset("rungalileo/ragbench", dataset)
     return ragbench
 
+def load_query_dataset(query_dataset):
+    global query_dataset_data
+    if query_dataset_data:
+        return query_dataset_data  
+    else:
+        query_dataset_data = load_dataset("rungalileo/ragbench", query_dataset)
+    return query_dataset_data
+
 def load_faiss(query_dataset):
     global index
     faiss_index_path = f"data_local/{query_dataset}_quantized.faiss"
@@ -135,7 +144,7 @@ def save_recent_question(question, metrics):
         "question": question,
         "metrics": metrics
     })
-
+    
     # Keep only the last 5 questions
     data["questions"]  = data["questions"][-5:]
 

evaluation.py

@@ -4,7 +4,8 @@ from sklearn.metrics import mean_squared_error, roc_auc_score
 from sklearn.feature_extraction.text import TfidfVectorizer
 from sklearn.metrics.pairwise import cosine_similarity
 
-from data_processing import load_ragbench
+from retrieval import query_dataset
+from data_processing import load_query_dataset
 
 global ground_truth_answer, ground_truth_metrics
 
@@ -92,7 +93,7 @@ def compute_rmse(predicted_values, ground_truth_values):
     return np.sqrt(mean_squared_error(ground_truth_values, predicted_values))
 
 def calculate_metrics(question, response, docs, time_taken):
-    data = load_ragbench()
+    data = load_query_dataset(query_dataset)
     ground_truth_answer = retrieve_ground_truths(question, data)  # Store the ground truth answer
 
     # Ensure ground_truth_answer is not empty before proceeding

retrieval.py

@@ -9,9 +9,11 @@ from sentence_transformers import CrossEncoder
 reranker = CrossEncoder("cross-encoder/ms-marco-MiniLM-L-6-v2")
 
 retrieved_docs = None
+query_dataset = 'hotpotqa'
 
 
 def retrieve_documents_hybrid(query, top_k=5):
+    global query_dataset
     query_dataset = find_query_dataset(query)
     
     with open( f"data_local/{query_dataset}_chunked_docs.json", "r") as f: