Update app.py

app.py

@@ -1,84 +1,250 @@
-from langchain.vectorstores import FAISS
+from langchain_community.vectorstores import FAISS
+from langchain_community.embeddings import HuggingFaceEmbeddings
+from langchain_community.document_loaders import PyPDFDirectoryLoader
+from langchain_community.llms import HuggingFaceEndpoint
+from langchain.chains import ConversationalRetrievalChain
 from langchain.chains import RetrievalQA
-from langchain.llms import HuggingFaceHub
 import gradio as gr
 import os
-from langchain.embeddings import HuggingFaceEmbeddings
-from langchain_experimental.agents.agent_toolkits.csv.base import create_csv_agent
-from langchain.document_loaders import PyPDFDirectoryLoader
+from pandasai import Agent
 from langchain.document_loaders.csv_loader import CSVLoader
 from langchain.text_splitter import RecursiveCharacterTextSplitter
+from langchain.memory import ConversationSummaryBufferMemory
 import io
 import contextlib
+import re
+import pandas as pd
+from transformers import AutoConfig
 
-
+config = AutoConfig.from_pretrained("config.json")
 embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
-vector_store= FAISS.load_local("vector_db/", embeddings)
+config = AutoConfig.from_pretrained("config.json")
+vector_store= FAISS.load_local("vector_db/", embeddings, allow_dangerous_deserialization=True)
 
 repo_id="mistralai/Mixtral-8x7B-Instruct-v0.1"
-llm = HuggingFaceHub(repo_id=repo_id, model_kwargs={"temperature": 0.01, "max_new_tokens": 2048})
+
+llm = HuggingFaceEndpoint(
+    repo_id = repo_id, 
+    temperature = 0.01, 
+    max_new_tokens = 4096,
+    verbose = True,
+    return_full_text = False 
+)
 
 retriever = vector_store.as_retriever(
     search_type="similarity",
-    search_kwargs={"k":3, "include_metadata": True})
+    search_kwargs={"k":5}
+)
+
+df=pd.read_csv('data/Gretel_Data.csv')
+averages = df.mean(numeric_only=True).to_dict()
+
+agent = Agent([df], config={"llm": llm, 'verbose':True})
+
+global unique_columns
+unique_columns = [
+        'Avg_Connected_UEs',
+        'PRB Util%',
+        'CA Activation Rate',
+        'DLRLCLayerDataVolume MB',
+        'DRB UL Data Volume MB',
+        'UPTP_Mbps',
+        'UPTP Mbps Num',
+        'UPTP Mbps Den',
+        'UL MAC Vol Scell Pct',
+        'DL MAC Vol Scell Pct',
+        'DL MAC Vol Scell MB',
+        'DL Volume',
+        'DL Data Vol MAC in MB',
+        'UL Throughput',
+        'MB_per_connected_UE'
+    ]
+
+global target_words
+target_words = ["Bandwidth", "Interference", "Call Quality", "Network", "Handover"]
+
+columns = []
 
-agent=create_csv_agent(llm,['data/Gretel_Data.csv','data/RAN_Data _T.csv'],verbose=True)
+column_avgs = {}
+
+global network_features
+network_features  = {
+    'Bandwidth': [
+        'Avg_Connected_UEs',
+        'PRB Util%',
+        'CA Activation Rate',
+        'DLRLCLayerDataVolume MB',
+        'DRB UL Data Volume MB',
+        'UPTP_Mbps',
+        'UPTP Mbps Num',
+        'UPTP Mbps Den',
+        'UL MAC Vol Scell Pct',
+        'DL MAC Vol Scell Pct',
+        'DL MAC Vol Scell MB',
+        'DL Volume',
+        'DL Data Vol MAC in MB',
+        'UL Throughput',
+        'MB_per_connected_UE'
+    ],
+    'Handover': [
+        'Avg_Connected_UEs',
+        'PRB Util%',
+        'CA Activation Rate',
+        'HO Failures',
+        'HO_fail_InterFreq',
+        'HO_fail_PCT_InterFreq',
+        'HO Failure%',
+        'HO Attempts',
+        'HO_att_InterFreq'
+    ],
+    'Network': [
+        'Avg_Connected_UEs',
+        'PRB Util%',
+        'CA Activation Rate',
+        'SIP DC%',
+        'RRC Setup Attempts',
+        'RRC Setup Failures',
+        'RRC Setup Failure% 5G',
+        'Combined RACH Failure%',
+        'Combined RACH Preambles',
+        'Combined RACH Failures',
+        'Interference Pwr',
+    ],
+    'Call Quality': [
+        'Avg_Connected_UEs',
+        'PRB Util%',
+        'CA Activation Rate',
+        'Avg_PUCCH_SINR',
+        'Avg CQI',
+        'SIP Calls with a Leg',
+        'SIP_SC_Total_MOU',
+        'SIP Dropped Calls',
+        'VoLTE_MOU',
+        'QCI 1 Bearer Drops',
+        'QCI 1 Bearer Releases',
+        'QCI 1 Bearer Drop%',
+        'Peak UE',
+        'DL Packet Loss Pct',
+        'UL Resid BLER PCT',
+        'Bearer Drops Voice',
+        'Bearer Releases Voice',
+        'Bearer Drop%',
+        'Call_Drops_Credit'
+    ],
+    'Interference': [
+        'Avg_Connected_UEs',
+        'PRB Util%',
+        'CA Activation Rate',
+        'Combined RACH Failure%',
+        'Interference Pwr'
+    ]
+}
 
 def echo(message, history):
   try:
-    qa=RetrievalQA.from_chain_type(llm=llm, retriever=retriever,return_source_documents=True)
-    message= "Your name is Clara. You are a senior telecom network engineer having access to troubleshooting tickets data and other technical and product documentation.Stick to the knowledge from these tickets. Ask clarification questions if needed.  "+message
-    result=qa({"query":message})
-    bold_answer= "<b>" + result['result'] + "</b>"
-    return bold_answer + "<br></br>" +'1. ' + str(result["source_documents"][0]) +"<br>" + '2. ' + str(result["source_documents"][1]) +  "<br>" + "3. " + str(result["source_documents"][2])
+    qa=RetrievalQA.from_chain_type(llm=llm, retriever=retriever, return_source_documents=True)
+    message= " <s> [INST] You are a senior telecom network engineer having access to troubleshooting tickets data and other technical and product documentation. Stick to the knowledge provided. Search through the product documentation pdfs first before scanning the tickets to generate the answer. Return only the helpful answer. Question:" + message + '[/INST]'
+    result= qa({"query":message})
+    answer= result['result']
+    for word in target_words:
+      if re.search(r'\b' + re.escape(word) + r'\b', answer, flags=re.IGNORECASE):
+        columns.extend(network_features.get(word, []))
+    unique_columns = list(set(columns))
+
+    for column in unique_columns:
+      column_avgs.update({column:averages.get(column, [])})
+
+    result_df = df[unique_columns].iloc[:25]
+
+    def highlight_rows(val, threshold):
+        if val > threshold:
+            return 'color: red; font-weight: bold'
+        elif val < threshold:
+            return 'color: green'
+        else:
+            return ''
+
+    styled_df = result_df.style
+
+    for key in column_avgs:
+      styled_df = styled_df.applymap(lambda x, k=key: highlight_rows(x, column_avgs[k]), subset=[f'{key}'])
+
+    gr.Dataframe(styled_df)
+    return (
+       "Answer: \n"
+       + '\n' + answer.strip() + '\n'
+       + '\n' + "Sources: \n"
+       + '\n' + '1. ' + result['source_documents'][0].metadata['source'] + '\n' + result['source_documents'][0].page_content +  "\n"
+       + '\n' + '2. ' + result['source_documents'][1].metadata['source'] + '\n' + result['source_documents'][1].page_content +  "\n"
+       + '\n' + "3. " + result['source_documents'][2].metadata['source'] + '\n' + result['source_documents'][2].page_content +  "\n"
+       + '\n' + "4. " + result['source_documents'][3].metadata['source'] + '\n' + result['source_documents'][3].page_content +  "\n"
+       + '\n' + "5. " + result['source_documents'][4].metadata['source'] + '\n' + result['source_documents'][4].page_content +  "\n",
+       styled_df
+       )
   except Exception as e:
     error_message = f"An error occurred: {e}"+str(e.with_traceback) + str(e.args)
+    return error_message, error_message
 
 def echo_agent(message, history):
-  message="There are 2 df's. If you find a KeyError check for the same in the other df." + "<br>" + message
   try:
-    with io.StringIO() as buffer:
-      with contextlib.redirect_stdout(buffer):
-        result= agent.run(message)
-      verbose_output = buffer.getvalue()
-      verbose_output = verbose_output.replace("\x1b[36;1m\x1b[1;3m", "")
-      verbose_output = verbose_output.replace("> ", "")
-      verbose_output = verbose_output.replace("", "")
-      verbose_output = verbose_output.replace("", "")
-      result= "<b>" + verbose_output + "<br>" + result + "</b>"
-    return result
+    response = agent.chat(message, output_type= 'text')
+    # explanation = agent.explain()
+
+    # result = "Answer: \n" + '\n' + response.str() + '\n' + '\n' + "Explanation: \n" + '\n' + explanation
+    
+    return response
   except Exception as e:
     error_message = f"An error occurred: {e}"+str(e.with_traceback) + str(e.args)
     return error_message
 
-demo=gr.ChatInterface(
-        fn=echo,
-        chatbot=gr.Chatbot(height=300, label="Hi I am Clara!", show_label=True),
-        textbox=gr.Textbox(placeholder="Ask me a question", container=True, autofocus=True, scale=7),
+demo_agent = gr.Blocks(
         title="Network Ticket Knowledge Management",
-        description="<span style='font-size: 16x;'>Welcome to Verizon Network Operations Center!! I am here to help the Verizon Field Operations team with technical queries & escalation. I am trained on 1000s of RAN, Backhaul, Core network & End user equipment trouble tickets. Ask me!!!&nbsp;☺</span>",
-        theme=gr.themes.Soft(),
-        examples=["wifi connected but no internet showing", "internet stopped working after primary link down", "internet stopped working link not shifted to secondary after primary link down"],
-        cache_examples=False,
-        retry_btn=None,
-        undo_btn="Delete Previous",
-        clear_btn="Clear",
-        stop_btn="Stop",
-    )
-
-
-demo1=gr.ChatInterface(
-        fn=echo_agent,
-        chatbot=gr.Chatbot(height=300, label="Hi I am Sam!", show_label=True),
-        textbox=gr.Textbox(placeholder="Ask me a question", container=True, autofocus=True, scale=7),
-        title="LLM Powered Agent",
-        description="<span style='font-size: 16x;'>Welcome to Verizon RAN Visualization & Analytics powered by GEN AI. I have access 100 of metrices generated by a RAN base station and can help in visualizing, correlating and generating insights, using power of Conversational AI &nbsp;☺</span>",
         theme=gr.themes.Soft(),
-        retry_btn=None,
-        undo_btn="Delete Previous",
-        clear_btn="Clear",
-        stop_btn="Stop",
-    )
-demo2=gr.TabbedInterface([demo,demo1],["RAG","AGENT"], title='INCEDO', theme=gr.themes.Soft())
-demo2.launch(share=True,debug=True,auth=("admin", "Sam&Clara"))
+)
+
+with demo_agent:
+
+  gr.Markdown(
+      '''
+      # <p style="text-align: center;">Network Ticket Knowledge Management</p>
+      Welcome to Verizon Network Operations Center. I am here to help the Field Operations team with technical queries & escalation.
+      '''
+      )
+
+  with gr.Tab('Clara'):
+    with gr.Row():
+      message = gr.Text(label="Input Query")
+
+    btn = gr.Button("Submit")
+
+    with gr.Row():
+      reply = gr.Text(label="RCA and MoP", autoscroll=False)
+
+    with gr.Accordion(label = "Metrics", open=False):
+          table = gr.Dataframe()
+
+    btn.click(echo, inputs=[message], outputs=[reply, table])
+
+    gr.Examples([
+        "Wi-Fi connected but no internet showing",
+        'What are the possible cause of router overheating ?',
+        "What are the possible causes of RAN getting disconnected frequently?",
+        "For the past week, are there any specific cell towers in Texas experiencing unusually high call failure rates or data latency?",
+        "What are the network problems faced by people living in the state of California?",
+        "I have an FWA connection and all devices except my iPhone have internet access via this FWA device. Can you suggest steps for resolution?",
+        "We're receiving reports of congested cell towers in Cleveland. Can you identify the specific cell towers experiencing overload and suggest any temporary network adjustments to alleviate the congestion?"
+        ],
+                inputs=[message]
+                )
+
+  with gr.Tab('Sam'):
+    with gr.Row():
+      message_agent = gr.Text(label="Input Query")
+    with gr.Row():
+      reply_agent = gr.Text(label="Answer")
+
+    btn2 = gr.Button("Submit")
+    btn2.click(echo_agent, inputs=[message_agent], outputs=[reply_agent])
+
 
+demo_agent.launch(share=True,debug=True,auth=("admin", "Sam&Clara"))