Update app.py

app.py

@@ -32,10 +32,19 @@ from pathlib import Path
 import torchaudio
 import numpy as np
 
-PYTORCH_USE_CUDA_DSA = 1
-CUDA_LAUNCH_BLOCKING = 1
-
-# Check if the token is already set in the environment variables
+# Neo4j imports
+from langchain.chains import GraphCypherQAChain
+from langchain_community.graphs import Neo4jGraph
+from langchain_community.document_loaders import HuggingFaceDatasetLoader
+from langchain_text_splitters import CharacterTextSplitter
+from langchain_experimental.graph_transformers import LLMGraphTransformer
+from langchain_core.prompts import ChatPromptTemplate
+from langchain_core.pydantic_v1 import BaseModel, Field
+from langchain_core.messages import AIMessage, HumanMessage
+from langchain_core.output_parsers import StrOutputParser
+from langchain_core.runnables import RunnableBranch, RunnableLambda, RunnableParallel, RunnablePassthrough
+
+# Pinecone setup
 hf_token = os.getenv("HF_TOKEN")
 if hf_token is None:
     print("Please set your Hugging Face token in the environment variables.")
@@ -45,7 +54,6 @@ else:
 logging.basicConfig(level=logging.DEBUG)
 
 embeddings = OpenAIEmbeddings(api_key=os.environ['OPENAI_API_KEY'])
-
 from pinecone import Pinecone
 pc = Pinecone(api_key=os.environ['PINECONE_API_KEY'])
 
@@ -61,167 +69,12 @@ conversational_memory = ConversationBufferWindowMemory(
     return_messages=True
 )
 
-# def get_current_time_and_date():
-#     now = datetime.now()
-#     return now.strftime("%Y-%m-%d %H:%M:%S")
-
-# current_time_and_date = get_current_time_and_date()
-
+# Prompt templates
 def get_current_date():
     return datetime.now().strftime("%B %d, %Y")
 
 current_date = get_current_date()
 
-def fetch_local_events():
-    api_key = os.environ['SERP_API']
-    url = f'https://serpapi.com/search.json?engine=google_events&q=Events+in+Birmingham&hl=en&gl=us&api_key={api_key}'
-    response = requests.get(url)
-    if response.status_code == 200:
-        events_results = response.json().get("events_results", [])
-        events_html = """
-        <h2 style="font-family: 'Georgia', serif; color: #ff0000; background-color: #f8f8f8; padding: 10px; border-radius: 10px;">Local Events</h2>
-        <style>
-            table {
-                font-family: 'Verdana', sans-serif;
-                color: #333;
-                border-collapse: collapse;
-                width: 100%;
-            }
-            th, td {
-                border: 1px solid #fff !important;
-                padding: 8px;
-            }
-            th {
-                background-color: #f2f2f2;
-                color: #333;
-                text-align: left;
-            }
-            tr:hover {
-                background-color: #f5f5f5;
-            }
-            .event-link {
-                color: #1E90FF;
-                text-decoration: none;
-            }
-            .event-link:hover {
-                text-decoration: underline;
-            }
-        </style>
-        <table>
-            <tr>
-                <th>Title</th>
-                <th>Date and Time</th>
-                <th>Location</th>
-            </tr>
-        """
-        for event in events_results:
-            title = event.get("title", "No title")
-            date_info = event.get("date", {})
-            date = f"{date_info.get('start_date', '')} {date_info.get('when', '')}".replace("{", "").replace("}", "")
-            location = event.get("address", "No location")
-            if isinstance(location, list):
-                location = " ".join(location)
-            location = location.replace("[", "").replace("]", "")
-            link = event.get("link", "#")
-            events_html += f"""
-            <tr>
-                <td><a class='event-link' href='{link}' target='_blank'>{title}</a></td>
-                <td>{date}</td>
-                <td>{location}</td>
-            </tr>
-            """
-        events_html += "</table>"
-        return events_html
-    else:
-        return "<p>Failed to fetch local events</p>"
-
-def fetch_local_weather():
-    try:
-        api_key = os.environ['WEATHER_API']
-        url = f'https://weather.visualcrossing.com/VisualCrossingWebServices/rest/services/timeline/birmingham?unitGroup=metric&include=events%2Calerts%2Chours%2Cdays%2Ccurrent&key={api_key}'
-        response = requests.get(url)
-        response.raise_for_status()
-        jsonData = response.json()
-        
-        current_conditions = jsonData.get("currentConditions", {})
-        temp_celsius = current_conditions.get("temp", "N/A")
-        
-        if temp_celsius != "N/A":
-            temp_fahrenheit = int((temp_celsius * 9/5) + 32)
-        else:
-            temp_fahrenheit = "N/A"
-            
-        condition = current_conditions.get("conditions", "N/A")
-        humidity = current_conditions.get("humidity", "N/A")
-
-        weather_html = f"""
-        <div class="weather-theme">
-            <h2 style="font-family: 'Georgia', serif; color: #ff0000; background-color: #f8f8f8; padding: 10px; border-radius: 10px;">Local Weather</h2>
-            <div class="weather-content">
-                <div class="weather-icon">
-                    <img src="https://www.weatherbit.io/static/img/icons/{get_weather_icon(condition)}.png" alt="{condition}" style="width: 100px; height: 100px;">
-                </div>
-                <div class="weather-details">
-                    <p style="font-family: 'Verdana', sans-serif; color: #333; font-size: 1.2em;">Temperature: {temp_fahrenheit}°F</p>
-                    <p style="font-family: 'Verdana', sans-serif; color: #333; font-size: 1.2em;">Condition: {condition}</p>
-                    <p style="font-family: 'Verdana', sans-serif; color: #333; font-size: 1.2em;">Humidity: {humidity}%</p>
-                </div>
-            </div>
-        </div>
-        <style>
-            .weather-theme {{
-                animation: backgroundAnimation 10s infinite alternate;
-                border-radius: 10px;
-                padding: 10px;
-                margin-bottom: 15px;
-                background: linear-gradient(45deg, #ffcc33, #ff6666, #ffcc33, #ff6666);
-                background-size: 400% 400%;
-                box-shadow: 0 4px 8px rgba(0, 0, 0, 0.1);
-                transition: box-shadow 0.3s ease, background-color 0.3s ease;
-            }}
-            .weather-theme:hover {{
-                box-shadow: 0 8px 16px rgba(0, 0, 0, 0.2);
-                background-position: 100% 100%;
-            }}
-            @keyframes backgroundAnimation {{
-                0% {{ background-position: 0% 50%; }}
-                100% {{ background-position: 100% 50%; }}
-            }}
-            .weather-content {{
-                display: flex;
-                align-items: center;
-            }}
-            .weather-icon {{
-                flex: 1;
-            }}
-            .weather-details {{
-                flex 3;
-            }}
-        </style>
-        """
-        return weather_html
-    except requests.exceptions.RequestException as e:
-        return f"<p>Failed to fetch local weather: {e}</p>"
-
-def get_weather_icon(condition):
-    condition_map = {
-        "Clear": "c01d",
-        "Partly Cloudy": "c02d",
-        "Cloudy": "c03d",
-        "Overcast": "c04d",
-        "Mist": "a01d",
-        "Patchy rain possible": "r01d",
-        "Light rain": "r02d",
-        "Moderate rain": "r03d",
-        "Heavy rain": "r04d",
-        "Snow": "s01d",
-        "Thunderstorm": "t01d",
-        "Fog": "a05d",
-    }
-    return condition_map.get(condition, "c04d")
-
-
-
 template1 = f"""As an expert concierge in Birmingham, Alabama, known for being a helpful and renowned guide, I am here to assist you on this sunny bright day of {current_date}. Given the current weather conditions and date, I have access to a plethora of information regarding events, places, and activities in Birmingham that can enhance your experience.
 If you have any questions or need recommendations, feel free to ask. I have a wealth of knowledge of perennial events in Birmingham and can provide detailed information to ensure you make the most of your time here. Remember, I am here to assist you in any way possible.
 Now, let me guide you through some of the exciting events happening today in Birmingham, Alabama:
@@ -261,56 +114,153 @@ Helpful Answer:"""
 QA_CHAIN_PROMPT_1 = PromptTemplate(input_variables=["context", "question"], template=template1)
 QA_CHAIN_PROMPT_2 = PromptTemplate(input_variables=["context", "question"], template=template2)
 
-def build_qa_chain(prompt_template):
-    qa_chain = RetrievalQA.from_chain_type(
-        llm=chat_model,
-        chain_type="stuff",
-        retriever=retriever,
-        chain_type_kwargs={"prompt": prompt_template}
-    )
-    tools = [
-        Tool(
-            name='Knowledge Base',
-            func=qa_chain,
-            description='Use this tool when answering general knowledge queries to get more information about the topic'
+# Neo4j setup
+graph = Neo4jGraph(
+    url="neo4j+s://d63baadd.databases.neo4j.io",
+    username="neo4j",
+    password="XCSXe1Jl_gjyJqoBGXDqY1UrfgDc4Z_RT5YGrxPAy-g"
+)
+
+dataset_name = "Pijush2023/birmindata07312024"
+page_content_column = 'events_description'
+loader = HuggingFaceDatasetLoader(dataset_name, page_content_column)
+data = loader.load()
+
+text_splitter = CharacterTextSplitter(chunk_size=100, chunk_overlap=50)
+documents = text_splitter.split_documents(data)
+
+embeddings = OpenAIEmbeddings(api_key=os.environ['OPENAI_API_KEY'])
+llm = ChatOpenAI(temperature=0, model='gpt-4o', api_key=os.environ['OPENAI_API_KEY'])
+
+llm_transformer = LLMGraphTransformer(llm=llm)
+graph_documents = llm_transformer.convert_to_graph_documents(documents)
+graph.add_graph_documents(graph_documents, baseEntityLabel=True, include_source=True)
+
+class Entities(BaseModel):
+    names: list[str] = Field(..., description="All the person, organization, or business entities that appear in the text")
+
+prompt = ChatPromptTemplate.from_messages([
+    ("system", "You are extracting organization and person entities from the text."),
+    ("human", "Use the given format to extract information from the following input: {question}"),
+])
+
+entity_chain = prompt | llm.with_structured_output(Entities)
+
+def remove_lucene_chars(input: str) -> str:
+    return input.translate(str.maketrans({"\\": r"\\", "+": r"\+", "-": r"\-", "&": r"\&", "|": r"\|", "!": r"\!", 
+                                          "(": r"\(", ")": r"\)", "{": r"\{", "}": r"\}", "[": r"\[", "]": r"\]", 
+                                          "^": r"\^", "~": r"\~", "*": r"\*", "?": r"\?", ":": r"\:", '"': r'\"', 
+                                          ";": r"\;", " ": r"\ "}))
+
+def generate_full_text_query(input: str) -> str:
+    full_text_query = ""
+    words = [el for el in remove_lucene_chars(input).split() if el]
+    for word in words[:-1]:
+        full_text_query += f" {word}~2 AND"
+    full_text_query += f" {words[-1]}~2"
+    return full_text_query.strip()
+
+def structured_retriever(question: str) -> str:
+    result = ""
+    entities = entity_chain.invoke({"question": question})
+    for entity in entities.names:
+        response = graph.query(
+            """CALL db.index.fulltext.queryNodes('entity', $query, {limit:2})
+            YIELD node,score
+            CALL {
+              WITH node
+              MATCH (node)-[r:!MENTIONS]->(neighbor)
+              RETURN node.id + ' - ' + type(r) + ' -> ' + neighbor.id AS output
+              UNION ALL
+              WITH node
+              MATCH (node)<-[r:!MENTIONS]-(neighbor)
+              RETURN neighbor.id + ' - ' + type(r) + ' -> ' +  node.id AS output
+            }
+            RETURN output LIMIT 50
+            """,
+            {"query": generate_full_text_query(entity)},
         )
-    ]
-    return qa_chain, tools
-
-def initialize_agent_with_prompt(prompt_template):
-    qa_chain, tools = build_qa_chain(prompt_template)
-    agent = initialize_agent(
-        agent='chat-conversational-react-description',
-        tools=tools,
-        llm=chat_model,
-        verbose=False,
-        max_iteration=5,
-        early_stopping_method='generate',
-        memory=conversational_memory
+        result += "\n".join([el['output'] for el in response])
+    return result
+
+def retriever_neo4j(question: str):
+    structured_data = structured_retriever(question)
+    return structured_data
+
+_template = """Given the following conversation and a follow-up question, rephrase the follow-up question to be a standalone question,
+in its original language.
+Chat History:
+{chat_history}
+Follow Up Input: {question}
+Standalone question:"""
+
+CONDENSE_QUESTION_PROMPT = PromptTemplate.from_template(_template)
+
+def _format_chat_history(chat_history: list[tuple[str, str]]) -> list:
+    buffer = []
+    for human, ai in chat_history:
+        buffer.append(HumanMessage(content=human))
+        buffer.append(AIMessage(content=ai))
+    return buffer
+
+_search_query = RunnableBranch(
+    (
+        RunnableLambda(lambda x: bool(x.get("chat_history"))).with_config(
+            run_name="HasChatHistoryCheck"
+        ),
+        RunnablePassthrough.assign(
+            chat_history=lambda x: _format_chat_history(x["chat_history"])
+        )
+        | CONDENSE_QUESTION_PROMPT
+        | ChatOpenAI(temperature=0, api_key=os.environ['OPENAI_API_KEY'])
+        | StrOutputParser(),
+    ),
+    RunnableLambda(lambda x : x["question"]),
+)
+
+template = """Answer the question based only on the following context:
+{context}
+
+Question: {question}
+Use natural language and be concise.
+Answer:"""
+
+prompt = ChatPromptTemplate.from_template(template)
+
+chain_neo4j = (
+    RunnableParallel(
+        {
+            "context": _search_query | retriever_neo4j,
+            "question": RunnablePassthrough(),
+        }
     )
-    return agent
+    | prompt
+    | llm
+    | StrOutputParser()
+)
 
-def generate_answer(message, choice):
-    logging.debug(f"generate_answer called with prompt_choice: {choice}")
-    
-    if choice == "Details":
-        agent = initialize_agent_with_prompt(QA_CHAIN_PROMPT_1)
-    elif choice == "Conversational":
-        agent = initialize_agent_with_prompt(QA_CHAIN_PROMPT_2)
+# Define a function to select between Pinecone and Neo4j
+def generate_answer(message, choice, retrieval_mode):
+    logging.debug(f"generate_answer called with choice: {choice} and retrieval_mode: {retrieval_mode}")
+
+    if retrieval_mode == "Vector":
+        qa_chain = build_qa_chain(QA_CHAIN_PROMPT_1 if choice == "Details" else QA_CHAIN_PROMPT_2)
+        agent = initialize_agent_with_prompt(QA_CHAIN_PROMPT_1 if choice == "Details" else QA_CHAIN_PROMPT_2)
+        response = agent(message)
+    elif retrieval_mode == "Knowledge-Graph":
+        response = chain_neo4j.invoke({"question": message})
     else:
-        logging.error(f"Invalid prompt_choice: {choice}. Defaulting to 'Conversational'")
-        agent = initialize_agent_with_prompt(QA_CHAIN_PROMPT_2)
-    response = agent(message)
+        response = "Invalid retrieval mode selected."
 
     addresses = extract_addresses(response['output'])
     return response['output'], addresses
 
-
-def bot(history, choice, tts_choice):
+# The rest of your Gradio code...
+def bot(history, choice, tts_choice, retrieval_mode):
     if not history:
         return history
     
-    response, addresses = generate_answer(history[-1][0], choice)
+    response, addresses = generate_answer(history[-1][0], choice, retrieval_mode)
     history[-1][1] = ""
 
     with concurrent.futures.ThreadPoolExecutor() as executor:
@@ -505,7 +455,6 @@ def show_map_if_details(history, choice):
     else:
         return gr.update(visible=False), ""
 
-
 def generate_audio_elevenlabs(text):
     XI_API_KEY = os.environ['ELEVENLABS_API']
     VOICE_ID = 'd9MIrwLnvDeH7aZb61E9'
@@ -676,11 +625,153 @@ def update_images():
     image_3 = generate_image(hardcoded_prompt_3)
     return image_1, image_2, image_3
 
+def fetch_local_events():
+    api_key = os.environ['SERP_API']
+    url = f'https://serpapi.com/search.json?engine=google_events&q=Events+in+Birmingham&hl=en&gl=us&api_key={api_key}'
+    response = requests.get(url)
+    if response.status_code == 200:
+        events_results = response.json().get("events_results", [])
+        events_html = """
+        <h2 style="font-family: 'Georgia', serif; color: #ff0000; background-color: #f8f8f8; padding: 10px; border-radius: 10px;">Local Events</h2>
+        <style>
+            table {
+                font-family: 'Verdana', sans-serif;
+                color: #333;
+                border-collapse: collapse;
+                width: 100%;
+            }
+            th, td {
+                border: 1px solid #fff !important;
+                padding: 8px;
+            }
+            th {
+                background-color: #f2f2f2;
+                color: #333;
+                text-align: left;
+            }
+            tr:hover {
+                background-color: #f5f5f5;
+            }
+            .event-link {
+                color: #1E90FF;
+                text-decoration: none;
+            }
+            .event-link:hover {
+                text-decoration: underline;
+            }
+        </style>
+        <table>
+            <tr>
+                <th>Title</th>
+                <th>Date and Time</th>
+                <th>Location</th>
+            </tr>
+        """
+        for event in events_results:
+            title = event.get("title", "No title")
+            date_info = event.get("date", {})
+            date = f"{date_info.get('start_date', '')} {date_info.get('when', '')}".replace("{", "").replace("}", "")
+            location = event.get("address", "No location")
+            if isinstance(location, list):
+                location = " ".join(location)
+            location = location.replace("[", "").replace("]", "")
+            link = event.get("link", "#")
+            events_html += f"""
+            <tr>
+                <td><a class='event-link' href='{link}' target='_blank'>{title}</a></td>
+                <td>{date}</td>
+                <td>{location}</td>
+            </tr>
+            """
+        events_html += "</table>"
+        return events_html
+    else:
+        return "<p>Failed to fetch local events</p>"
 
+def get_weather_icon(condition):
+    condition_map = {
+        "Clear": "c01d",
+        "Partly Cloudy": "c02d",
+        "Cloudy": "c03d",
+        "Overcast": "c04d",
+        "Mist": "a01d",
+        "Patchy rain possible": "r01d",
+        "Light rain": "r02d",
+        "Moderate rain": "r03d",
+        "Heavy rain": "r04d",
+        "Snow": "s01d",
+        "Thunderstorm": "t01d",
+        "Fog": "a05d",
+    }
+    return condition_map.get(condition, "c04d")
 
-def clear_textbox():
-    return ""
+def fetch_local_weather():
+    try:
+        api_key = os.environ['WEATHER_API']
+        url = f'https://weather.visualcrossing.com/VisualCrossingWebServices/rest/services/timeline/birmingham?unitGroup=metric&include=events%2Calerts%2Chours%2Cdays%2Ccurrent&key={api_key}'
+        response = requests.get(url)
+        response.raise_for_status()
+        jsonData = response.json()
+        
+        current_conditions = jsonData.get("currentConditions", {})
+        temp_celsius = current_conditions.get("temp", "N/A")
+        
+        if temp_celsius != "N/A":
+            temp_fahrenheit = int((temp_celsius * 9/5) + 32)
+        else:
+            temp_fahrenheit = "N/A"
+            
+        condition = current_conditions.get("conditions", "N/A")
+        humidity = current_conditions.get("humidity", "N/A")
 
+        weather_html = f"""
+        <div class="weather-theme">
+            <h2 style="font-family: 'Georgia', serif; color: #ff0000; background-color: #f8f8f8; padding: 10px; border-radius: 10px;">Local Weather</h2>
+            <div class="weather-content">
+                <div class="weather-icon">
+                    <img src="https://www.weatherbit.io/static/img/icons/{get_weather_icon(condition)}.png" alt="{condition}" style="width: 100px; height: 100px;">
+                </div>
+                <div class="weather-details">
+                    <p style="font-family: 'Verdana', sans-serif; color: #333; font-size: 1.2em;">Temperature: {temp_fahrenheit}°F</p>
+                    <p style="font-family: 'Verdana', sans-serif; color: #333; font-size: 1.2em;">Condition: {condition}</p>
+                    <p style="font-family: 'Verdana', sans-serif; color: #333; font-size: 1.2em;">Humidity: {humidity}%</p>
+                </div>
+            </div>
+        </div>
+        <style>
+            .weather-theme {{
+                animation: backgroundAnimation 10s infinite alternate;
+                border-radius: 10px;
+                padding: 10px;
+                margin-bottom: 15px;
+                background: linear-gradient(45deg, #ffcc33, #ff6666, #ffcc33, #ff6666);
+                background-size: 400% 400%;
+                box-shadow: 0 4px 8px rgba(0, 0, 0, 0.1);
+                transition: box-shadow 0.3s ease, background-color 0.3s ease;
+            }}
+            .weather-theme:hover {{
+                box-shadow: 0 8px 16px rgba(0, 0, 0, 0.2);
+                background-position: 100% 100%;
+            }}
+            @keyframes backgroundAnimation {{
+                0% {{ background-position: 0% 50%; }}
+                100% {{ background-position: 100% 50%; }}
+            }}
+            .weather-content {{
+                display: flex;
+                align-items: center;
+            }}
+            .weather-icon {{
+                flex: 1;
+            }}
+            .weather-details {{
+                flex 3;
+            }}
+        </style>
+        """
+        return weather_html
+    except requests.exceptions.RequestException as e:
+        return f"<p>Failed to fetch local weather: {e}</p>"
 
 with gr.Blocks(theme='Pijush2023/scikit-learn-pijush') as demo:
     with gr.Row():
@@ -689,6 +780,7 @@ with gr.Blocks(theme='Pijush2023/scikit-learn-pijush') as demo:
 
             chatbot = gr.Chatbot([], elem_id="RADAR:Channel 94.1", bubble_full_width=False)
             choice = gr.Radio(label="Select Style", choices=["Details", "Conversational"], value="Conversational")
+            retrieval_mode = gr.Radio(label="Retrieval Mode", choices=["Vector", "Knowledge-Graph"], value="Vector")
 
             gr.Markdown("<h1 style='color: red;'>Talk to RADAR</h1>", elem_id="voice-markdown")
 
@@ -713,7 +805,7 @@ with gr.Blocks(theme='Pijush2023/scikit-learn-pijush') as demo:
             retriever_sequence = (
                 retriever_button.click(fn=stop_audio, inputs=[], outputs=[audio_output],api_name="Ask_Retriever")
                 .then(fn=add_message, inputs=[chatbot, chat_input], outputs=[chatbot, chat_input],api_name="voice_query")
-                .then(fn=bot, inputs=[chatbot, choice, tts_choice], outputs=[chatbot, audio_output],api_name="generate_voice_response" )
+                .then(fn=bot, inputs=[chatbot, choice, tts_choice, retrieval_mode], outputs=[chatbot, audio_output],api_name="generate_voice_response" )
                 .then(fn=show_map_if_details, inputs=[chatbot, choice], outputs=[location_output, location_output], api_name="map_finder")
                 .then(fn=clear_textbox, inputs=[], outputs=[chat_input])
             )
@@ -721,7 +813,7 @@ with gr.Blocks(theme='Pijush2023/scikit-learn-pijush') as demo:
             # Link the "Enter" key (submit event) to the same sequence of actions
             chat_input.submit(fn=stop_audio, inputs=[], outputs=[audio_output])
             chat_input.submit(fn=add_message, inputs=[chatbot, chat_input], outputs=[chatbot, chat_input],api_name="voice_query").then(
-                fn=bot, inputs=[chatbot, choice, tts_choice], outputs=[chatbot, audio_output], api_name="generate_voice_response"
+                fn=bot, inputs=[chatbot, choice, tts_choice, retrieval_mode], outputs=[chatbot, audio_output], api_name="generate_voice_response"
             ).then(
                 fn=show_map_if_details, inputs=[chatbot, choice], outputs=[location_output, location_output], api_name="map_finder"
             ).then(
@@ -746,5 +838,3 @@ with gr.Blocks(theme='Pijush2023/scikit-learn-pijush') as demo:
 
 demo.queue()
 demo.launch(share=True)
-
-