analyzer.py

import panel as pn
from langchain_community.llms import OpenAI
from langchain_experimental.agents.agent_toolkits.csv.base import create_csv_agent
import pandas as pd
import plotly.express as px
from dotenv import load_dotenv
import warnings
import io
import os

warnings.filterwarnings("ignore")
pn.extension('plotly', 'tabulator', comms="vscode")
load_dotenv()
print(os.environ["OPENAI_API_KEY"])

file_name = "data/data.csv"
data = pd.read_csv(file_name).drop(columns=['id'])

plot_pane = pn.pane.Plotly(sizing_mode="stretch_width")

file_input = pn.widgets.FileInput()
text_input = pn.widgets.TextInput(
    name="Question",
    placeholder="Ask a question from the CSV",
    sizing_mode="scale_width"
)

ask_button = pn.widgets.Button(
    name="Ask",
    button_type="primary",
    height=60
)

load_button = pn.widgets.Button(
    name="Load",
    button_type="primary",
)

plot_button = pn.widgets.Button(
    name="Plot",
    button_type="primary",
)

chat_box = pn.chat.ChatInterface(
    user="Asker",
    avatar="?",
    callback_user="Counter",
)
def load_page(data, file_name):
    target = data.columns[-1]
    
    print("load_page", data)

    yaxis = pn.widgets.Select(
        name='Y axis',
        options=list(data.columns),
        value=list(data._get_numeric_data().columns)[0],
        disabled_options=list(set(data.columns) - set(data._get_numeric_data().columns))
    )

    xaxis = pn.widgets.Select(
        name='X axis',
        options=list(data.columns),
        value=list(data._get_numeric_data().columns)[1],
        disabled_options=list(set(data.columns) - set(data._get_numeric_data().columns))
    )

    plot = px.scatter(
        data,
        x=list(data._get_numeric_data().columns)[0],
        y=list(data._get_numeric_data().columns)[1],
        color=target,
    )

    table = pn.widgets.Tabulator(data, page_size=20)

    agent = create_csv_agent(
        OpenAI(model="gpt-3.5-turbo-instruct", temperature=0),
        file_name,
        verbose=True,
        return_intermediate_steps=True,
        allow_dangerous_code=True
    )

    return target, yaxis, xaxis, plot, table, agent

target, yaxis, xaxis, plot, table, agent = load_page(data, file_name)

template = pn.template.FastListTemplate(
    title="CSV Analyzer",
    sidebar=[
        pn.Row(pn.pane.Markdown("# Analyze your CSVs")),
        pn.Row(pn.pane.Markdown("## Settings")),
        pn.Row(file_input, load_button),
        pn.Row(yaxis),
        pn.Row(xaxis),
        pn.Row(plot_button),
        pn.Row(plot_pane)
    ],
    main=[
        pn.Row(pn.Column(table), height=200),
        pn.Row(pn.Column(text_input), pn.Column(ask_button)),
        pn.Row(pn.Column(chat_box, sizing_mode='scale_width', scroll=True))
    ],
    sidebar_width=420,
    accent_base_color="#88d8b0",
    header_background="#88d8b0"
)

def reloader(event):
    plot_pane = px.scatter(data, x=xaxis.value, y=yaxis.value, color=target)
    template.sidebar[6][0].object = plot_pane

plot_button.on_click(reloader)

def parse_file_input(event):
    global data, file_name

    value = file_input.value
    bytes_io = io.BytesIO(value)
    data = pd.read_csv(bytes_io)
    file_name = file_input.file_name

    data.to_csv(file_name, index=False)

    global target, yaxis, xaxis, plot, table, agent
    target, yaxis, xaxis, plot, table, agent = load_page(data, file_name)

    template.main[0][0] = table
    template.main[3][0] = yaxis
    template.main[4][0] = xaxis
    template.main[6][0] = plot

load_button.on_click(parse_file_input)

def ask(event):
    query = text_input.value
    chat_box.append({"User": query})
    response = agent({"input": query})
    chat_box.append({"Thought Process": [x[0].log for x in response["intermediate_steps"]]})
    chat_box.append({"Bot": response["output"]})

ask_button.on_click(ask)

template.show(open=False, address='0.0.0.0', port=3000, websocket_origin='127.0.0.1:3000') # websocket_origin=url



# agent = create_csv_agent(
#     OpenAI(model="gpt-3.5-turbo-instruct", temperature=0), 
#     file_name, 
#     verbose=True,
#     return_intermediate_steps=True
# )

# agent.run("What is the average price of the products?")