\nYou will need the following to run scripts in this article:\n<\/p>\n
- \n
- A GridDB cloud account. Sign up for GridDB cloud<\/a> and complete configuration settings.<\/li>\n
- OpenAI API Key<\/a>. You can use any other LLM provider, but you will need to update the scripts in this article slightly.<\/li>\n<\/ol>\n
Installing and Importing Required Libraries<\/h2>\n
\nThe following script installs and imports the required libraries for this application:\n<\/p>\n
\n!pip install langchain\r\n!pip install langchain-core\r\n!pip install langchain-community\r\n!pip install langgraph\r\n!pip install langchain_huggingface\r\n!pip install tabulate\r\n!pip uninstall -y pydantic\r\n!pip install --no-cache-dir "pydantic>=2.11,<3"<\/code><\/pre>\n<\/div>\n\nimport pandas as pd\r\nimport json\r\nimport datetime as dt\r\nimport base64\r\nimport requests\r\nimport numpy as np\r\nfrom pathlib import Path\r\nimport matplotlib\r\nmatplotlib.use("Agg") # safe, consistent backend\r\nimport matplotlib.pyplot as plt\r\nfrom typing_extensions import Annotated\r\nfrom operator import add # used as list reducer\r\nfrom typing import TypedDict, List, Dict\r\nfrom pydantic import BaseModel, Field\r\nfrom IPython.display import Image, display\r\n\r\nfrom langchain_core.prompts import ChatPromptTemplate\r\nfrom langchain_core.tools import tool\r\nfrom langchain_openai import ChatOpenAI, OpenAI\r\nfrom langgraph.graph import START, END, StateGraph\r\nfrom langchain_core.messages import HumanMessage\r\nfrom langchain_experimental.agents import create_pandas_dataframe_agent\r\nfrom langgraph.prebuilt import create_react_agent\r\nfrom langchain.agents.agent_types import AgentType<\/code><\/pre>\n<\/div>\nImporting the Dataset<\/h2>\n
\nWe will insert the Titanic dataset<\/a> into GridDB and create visualizations using this data. The following script imports the data into a Pandas dataframe.\n<\/p>\n
\ndataset = pd.read_csv("https:\/\/raw.githubusercontent.com\/datasciencedojo\/datasets\/refs\/heads\/master\/titanic.csv",\r\n encoding = 'utf-8')\r\ndataset.head()<\/code><\/pre>\n<\/div>\nOutput:<\/strong>
\n![\"\"](\"\/wp-content\/uploads\/2026\/07\/img1-titanic-dataset-header.png\")
<\/a><\/p>\nEstablishing a Connection with GridDB Cloud<\/h2>\n
\nTo establish a connection with GridDB, replace your credentials in the following script and run it.\n<\/p>\n
\nusername = "USER_NAME"\r\npassword = "PASSWORD"\r\nbase_url = "GRIDDB_CLOUD_URL"\r\n\r\nurl = f"{base_url}\/checkConnection"\r\n\r\ncredentials = f"{username}:{password}"\r\nencoded_credentials = base64.b64encode(credentials.encode()).decode()\r\n\r\nheaders = {\r\n 'Content-Type': 'application\/json', # Added this header to specify JSON content\r\n 'Authorization': f'Basic {encoded_credentials}',\r\n 'User-Agent': 'PostmanRuntime\/7.29.0'\r\n}\r\n\r\nresponse = requests.get(url, headers=headers)\r\n\r\nprint(response.status_code)\r\nprint(response.text)<\/code><\/pre>\n<\/div>\nOutput:<\/strong><\/p>\n
\n200<\/code><\/pre>\n<\/div>\n\nIf you see the above message, you have successfully connected with GridDB cloud.\n<\/p>\n
Inserting Data in GridDB Cloud Dataset<\/h2>\n
\nTo insert data in GridDB, you first need to map your dataset types to GridDB dataset types and then create a GridDB container.\n<\/p>\n
Creating a Container for the Titanic Dataset in GridDB<\/h3>\n
\nThe following script maps your dataset column types to GridDB column types.\n<\/p>\n
\ndataset.insert(0, "SerialNo", dataset.index + 1)\r\ndataset.columns.name = None\r\n# Mapping pandas dtypes to GridDB types\r\ntype_mapping = {\r\n "int64": "LONG",\r\n "float64": "DOUBLE",\r\n "bool": "BOOL",\r\n 'datetime64': "TIMESTAMP",\r\n "object": "STRING",\r\n "category": "STRING",\r\n}\r\n\r\n# Generate the columns part of the payload dynamically\r\ncolumns = []\r\nfor col, dtype in dataset.dtypes.items():\r\n griddb_type = type_mapping.get(str(dtype), "STRING") # Default to STRING if unknown\r\n columns.append({\r\n "name": col,\r\n "type": griddb_type\r\n })\r\n\r\nprint(columns)<\/code><\/pre>\n<\/div>\n\n[{'name': 'SerialNo', 'type': 'LONG'},\r\n{'name': 'PassengerId', 'type': 'LONG'},\r\n{'name': 'Survived', 'type': 'LONG'},\r\n{'name': 'Pclass', 'type': 'LONG'},\r\n{'name': 'Name', 'type': 'STRING'},\r\n{'name': 'Sex', 'type': 'STRING'},\r\n{'name': 'Age', 'type': 'DOUBLE'},\r\n{'name': 'SibSp', 'type': 'LONG'},\r\n{'name': 'Parch', 'type': 'LONG'},\r\n{'name': 'Ticket', 'type': 'STRING'},\r\n{'name': 'Fare', 'type': 'DOUBLE'},\r\n{'name': 'Cabin', 'type': 'STRING'},\r\n{'name': 'Embarked', 'type': 'STRING'}]<\/code><\/pre>\n<\/div>\n\nNext, we will create a collection type container
titanic_db<\/code> in our GridDB cloud database.\n<\/p>\n\nurl = f"{base_url}\/containers"\r\ncontainer_name = "titanic_db"\r\n# Create the payload for the POST request\r\npayload = json.dumps({\r\n "container_name": container_name,\r\n "container_type": "COLLECTION",\r\n "rowkey": True, # Assuming the first column as rowkey\r\n "columns": columns\r\n})\r\n\r\n\r\n# Make the POST request to create the container\r\nresponse = requests.post(url, headers=headers, data=payload)\r\n\r\n# Print the response\r\nprint(f"Status Code: {response.status_code}")<\/code><\/pre>\n<\/div>\nOutput:<\/strong><\/p>\n
\nStatus Code: 201<\/code><\/pre>\n<\/div>\nInserting Titanic Dataset in GridDB<\/h3>\n
\nNext, we will iterate through the rows in our dataset, create a JSON payload containing the data, and will insert the data into the container we created in the previous section.\n<\/p>\n
\nurl = f"{base_url}\/containers\/{container_name}\/rows"\r\n# Convert dataset to list of lists (row-wise) with proper formatting\r\n\r\ndef format_row(row):\r\n formatted = []\r\n for item in row:\r\n if pd.isna(item):\r\n formatted.append(None) # Convert NaN to None\r\n elif isinstance(item, bool):\r\n formatted.append(str(item).lower()) # Convert True\/False to true\/false\r\n elif isinstance(item, (int, float)):\r\n formatted.append(item) # Keep integers and floats as they are\r\n else:\r\n formatted.append(str(item)) # Convert other types to string\r\n return formatted\r\n\r\n# Prepare rows with correct formatting\r\nrows = [format_row(row) for row in dataset.values.tolist()]\r\n\r\n# Create payload as a JSON string\r\npayload = json.dumps(rows)\r\n\r\n# Make the PUT request to add the rows to the container\r\nresponse = requests.put(url, headers=headers, data=payload)\r\n\r\n# Print the response\r\nprint(f"Status Code: {response.status_code}")\r\nprint(f"Response Text: {response.text}")<\/code><\/pre>\n<\/div>\nOutput:<\/strong><\/p>\n
\nStatus Code: 200\r\nResponse Text: {"count":891}<\/code><\/pre>\n<\/div>\n\nThe above output shows that the data has been successfully inserted into GridDB.\n<\/p>\n
\nNext, we will see how to retrieve data from GridDB and plot visualizations using it.\n<\/p>\n
Visualizing GridDB Results Using OpenAI and ReAct Agent<\/h2>\n
\nThe following script reads data from GridDB and inserts it in a Pandas dataframe.\n<\/p>\n
\ncontainer_name = "titanic_db"\r\n\r\nurl = f"{base_url}\/containers\/{container_name}\/rows"\r\n\r\n# Define the payload for the query\r\npayload = json.dumps({\r\n "offset": 0, # Start from the first row\r\n "limit": 10000, # Limit the number of rows returned\r\n "condition": "", # No filtering condition (you can customize it)\r\n "sort": "" # No sorting (you can customize it)\r\n})\r\n\r\n# Make the POST request to read data from the container\r\nresponse = requests.post(url, headers=headers, data=payload)\r\n\r\n# Check response status and print output\r\nprint(f"Status Code: {response.status_code}")\r\nif response.status_code == 200:\r\n try:\r\n data = response.json()\r\n print("Data retrieved successfully!")\r\n\r\n # Convert the response to a DataFrame\r\n rows = data.get("rows", [])\r\n titanic_dataset = pd.DataFrame(rows, columns=[col for col in dataset.columns])\r\n\r\n except json.JSONDecodeError:\r\n print("Error: Failed to decode JSON response.")\r\nelse:\r\n print(f"Error: Failed to query data from the container. Response: {response.text}")\r\n\r\nprint(titanic_dataset.shape)\r\ntitanic_dataset.head()<\/code><\/pre>\n<\/div>\nOutput:<\/strong>
\n![\"\"](\"\/wp-content\/uploads\/2026\/07\/img2-data-retrieved-from-griddb.png\")
<\/a><\/p>\n\nLet’s try to plot the average for the passengers who survived and those who didn’t. We will use these values to verify the result from our ReAct agent.\n<\/p>\n
\navg = titanic_dataset.groupby('Survived')['Fare'].mean().round(2)\r\nprint(avg)<\/code><\/pre>\n<\/div>\nOutput:<\/strong><\/p>\n
\nSurvived\r\n0 22.12\r\n1 48.40\r\nName: Fare, dtype: float64<\/code><\/pre>\n<\/div>\nCreating a LangGraph ReAct Agent for Data Visualization<\/h3>\n
\nTo plot visualizations, we will create a LangGraph ReAct<\/a> agent with two tools:
df_answer<\/code> andsave_plot<\/code>.\n<\/p>\n\nThe
df_answer<\/code> tool will use thecreate_pandas_dataframe_agent<\/code><\/a> to retrieve information from the database, including the plot, if any. Thesave_plot<\/code> tool saves the plot to the local drive.\n<\/p>\n\nThe following script defines the agent’s state and the large language model (OpenAI GPT-4o in this case) we will use to answer the user’s question.\n<\/p>\n
\nclass State(TypedDict):\r\n question: str\r\n answer: str\r\n plots: Annotated[List[Dict[str, str]], add]\r\n\r\n\r\napi_key = "YOUR_OPENAI_API_KEY"\r\n\r\nllm = ChatOpenAI(model="gpt-4o",\r\n api_key=api_key,\r\n temperature = 0)<\/code><\/pre>\n<\/div>\n\nNext, we define the
create_pandas_dataframe_agent<\/code> function that returns information from the Pandas dataframe retrieved from GridDB.
\nWe also define thedf_answer<\/code> tool that calls thecreate_pandas_dataframe_agent.<\/code>\n<\/p>\n\ndf_agent = create_pandas_dataframe_agent(llm,\r\n titanic_dataset,\r\n verbose=True,\r\n agent_type=AgentType.ZERO_SHOT_REACT_DESCRIPTION,\r\n allow_dangerous_code=True)\r\n\r\n\r\n_LAST = {"fig": None} # keep a handle to the last real figure\r\n\r\n@tool("df_answer")\r\ndef df_answer(question: str) -> str:\r\n """\r\n Use the pandas DataFrame agent to compute\/plot.\r\n IMPORTANT: do NOT call plt.show() or plt.close() in the generated code.\r\n """\r\n res = df_agent.invoke(\r\n question\r\n )\r\n\r\n # CAPTURE whichever figure the agent actually created\r\n fig_nums = plt.get_fignums() # existing figures in this process\r\n if fig_nums:\r\n _LAST["fig"] = plt.figure(fig_nums[-1]) # latest real figure\r\n return res["output"]<\/code><\/pre>\n<\/div>\n\nThe following script defines the
save_plot<\/code> tool that saves the plot generated by thedf_answer<\/code> tool.\n<\/p>\n\n\r\n@tool("save_plot")\r\ndef save_plot(filename: str = "plot.png", dpi: int = 200, close: bool = True) -> str:\r\n """\r\n Save the most recent existing Matplotlib figure (not an empty gcf()).\r\n Returns {"plot": {"name": ..., "path": ...}} or {"error": ...}.\r\n """\r\n Path("plots").mkdir(exist_ok=True)\r\n fig = _LAST.get("fig")\r\n\r\n # Fallback: grab last live figure if we didn't capture yet\r\n if fig is None:\r\n nums = plt.get_fignums()\r\n if not nums:\r\n return json.dumps({"error": "no_figure", "message": "No active figure to save."})\r\n fig = plt.figure(nums[-1])\r\n\r\n # Render + save\r\n fig.tight_layout()\r\n try:\r\n fig.canvas.draw() # ensure render\r\n except Exception:\r\n pass\r\n\r\n out = Path("plots") \/ filename\r\n fig.savefig(out, dpi=dpi, bbox_inches="tight")\r\n\r\n if close:\r\n plt.close(fig) # avoid accumulating figures\r\n\r\n return json.dumps({"plot": {"name": filename, "path": str(out.resolve())}})<\/code><\/pre>\n<\/div>\n\nFinally we define the ReAct agent using the LLM and the tool we just defined.\n<\/p>\n
\n\r\nSYSTEM = """\r\nYou work over a Titanic pandas DataFrame.\r\n- To compute answers or create charts, call `df_answer(question=...)`.\r\n- If a plot should be saved, call `save_plot(filename=..., dpi=200)`.\r\n- Keep text concise. If you saved a plot, you may echo the absolute path.\r\n"""\r\n\r\nreact = create_react_agent(\r\n llm,\r\n tools=[df_answer, save_plot],\r\n prompt=SYSTEM,\r\n)<\/code><\/pre>\n<\/div>\n\nThe following script creates our final graph object.\n<\/p>\n
\ndef run_react(state: State) -> State:\r\n out = react.invoke({"messages": [("user", state["question"])]})\r\n\r\n msgs = out["messages"]\r\n final_text = msgs[-1].content\r\n\r\n new_plots = []\r\n for m in msgs:\r\n # Tool messages include the tool's return in `content`\r\n try:\r\n data = json.loads(getattr(m, "content", "") or "{}")\r\n except Exception:\r\n data = None\r\n if isinstance(data, dict) and "plot" in data:\r\n new_plots.append(data["plot"]) # {"name": "...", "path": "..."}\r\n\r\n return {"answer": final_text, "plots": new_plots}\r\n\r\ngraph_builder = StateGraph(State)\r\ngraph_builder.add_node("ask_question", run_react)\r\ngraph_builder.add_edge(START, "ask_question")\r\ngraph_builder.add_edge("ask_question", END)\r\ngraph = graph_builder.compile()\r\ndisplay(Image(graph.get_graph(xray=True).draw_mermaid_png()))<\/code><\/pre>\n<\/div>\nOutput:<\/strong>
\n![\"\"](\"\/wp-content\/uploads\/2026\/07\/img3-react-agent-for-data-visualization.png\")
<\/a><\/p>\n\nThe above output shows the flow of the graph. The user’s question is passed to the ReAct agent, which decides which tools it requires to answer the user’s query.\n<\/p>\n
Testing the Agent & Generating Responses<\/h3>\n
\nLet’s test the agent. We will first ask a simple question that doesn’t require saving or plotting a graph.\n<\/p>\n
\n# A) plain Q&A\r\ns = graph.invoke({"question": "What is the average Fare for passengers that survive and those who did not?"})\r\nprint(f"\\nFinal Answer: {s['answer']}")<\/code><\/pre>\n<\/div>\nOutput:<\/strong>
\n![\"\"](\"\/wp-content\/uploads\/2026\/07\/img4-response-from-text-only-agent.png\")
<\/a><\/p>\n\nThe output displays the average fares for passengers who survived and those who did not. Note that these values are identical to the ones we retrieved earlier by executing a direct operation on the Pandas dataframe.\n<\/p>\n
\nNext, we will request our agent to plot a chart using the results.\n<\/p>\n
\n# B) plot + save (the agent will call save_plot internally)\r\ns = graph.invoke({\r\n "question": ("Plot a bar chart of average Fare for passengers that survive and those who did not?"),\r\n})\r\nprint(f"\\nFinal Answer: {s['answer']}")\r\nprint("plots so far:", s["plots"])<\/code><\/pre>\n<\/div>\nOutput:<\/strong>
\n![\"\"](\"\/wp-content\/uploads\/2026\/07\/img5-response-from-data-visualization-agent.png\")
<\/a><\/p>\n\nThe output shows that the agent saved the plot and also returned its location in the output.\n<\/p>\n
\nIf you open the plot, you can see the average fare by survival rate plotted in the form of a bar chart.\n<\/p>\n
- OpenAI API Key<\/a>. You can use any other LLM provider, but you will need to update the scripts in this article slightly.<\/li>\n<\/ol>\n
![\"\"](\"\/wp-content\/uploads\/2026\/07\/img6-final_plot.png\")
<\/a><\/p>\n