app.py

'''
参考： https://github.com/shroominic/codeinterpreter-api

1. 可以存在本地，然后再调出来。 working. 
1. 可以直接在内存中读出图片。
'''
# TODO：如何在内存中读取文件。

import matplotlib as mpl
from codeinterpreterapi import CodeInterpreterSession, File
import streamlit as st
from codeinterpreterapi import CodeInterpreterSession
import openai
import os
import matplotlib.pyplot as plt
import pandas as pd
from io import StringIO
import csv
import tempfile
from tempfile import NamedTemporaryFile
import pathlib
from pathlib import Path
import matplotlib
from matplotlib.font_manager import FontProperties
import seaborn as sns

os.environ["OPENAI_API_KEY"] = os.environ['user_token']
openai.api_key = os.environ['user_token']
os.environ["VERBOSE"] = "True"  # 可以看到具体的错误？

# 设置中文字体。
myfont = FontProperties(fname='YaHei.ttf')  # NOTE: 注意这里的云服务器路径格式。
mpl.rcParams['font.family'] = ['myfont']
sns.set(font='myfont')
plt.rcParams['font.sans-serif'] = ['myfont']
plt.rcParams['font.family'] = 'sans-serif'
# plt.title("这个是数据分布图", fontsize=12, fontproperties=myfont)
# plt.xlabel('数据A', fontproperties=myfont)


# # #* 如果碰到接口问题，可以启用如下设置。
# openai.proxy = {
#             "http": "http://127.0.0.1:7890",
#             "https": "http://127.0.0.1:7890"
#         }


# st.title("ChatGPT-like clone")
st.title("大语言模型商业数据分析中心")
st.subheader("Business Data Analytics Based Upon LLM")
uploaded_file = st.file_uploader(
    "Choose a file", type=(["csv", "txt", "xlsx", "xls"]))
# uploaded_file = st.file_uploader("选择一个文件", type=(["csv","txt","xlsx","xls"]))
# st.write(uploaded_file)
if uploaded_file is not None:
    # csv_file = csv.reader(uploaded_file)
    csv_file = pd.read_csv(uploaded_file)
    st.write(csv_file[:5])  # 这里只是显示文件，后面需要定位文件所在的绝对路径。

    uploaded_file_name = "File_provided"
    temp_dir = tempfile.TemporaryDirectory()
    # ! working.
    uploaded_file_path = pathlib.Path(temp_dir.name) / uploaded_file_name
    with open(uploaded_file_path, 'wb') as output_temporary_file:
        # output_temporary_file.write(uploaded_file.read())
        # ! 必须用这种格式读入内容，然后才可以写入temporary文件夹中。
        output_temporary_file.write(uploaded_file.getvalue())
        # st.write(uploaded_file_path) #* 可以查看文件是否真实存在，然后是否可以

# how to read data inside streamlit.
#     # files = pd.read_csv(uploaded_file)
#     bytes_data = uploaded_file.getvalue()
#     # st.write(bytes_data)

#     # To convert to a string based IO:
#     stringio = StringIO(uploaded_file.getvalue().decode("utf-8"))
#     # st.write(stringio)

#     # To read file as string:
#     string_data = stringio.read()
#     # st.write(string_data)

#     # Can be used wherever a "file-like" object is accepted:
#     # dataframe = pd.read_csv(uploaded_file)
#     files = pd.read_csv(uploaded_file, encoding='utf-8')

# openai.api_key = st.secrets["OPENAI_API_KEY"]


async def main():
    if "openai_model" not in st.session_state:
        # st.session_state["openai_model"] = "gpt-3.5-turbo"
        # NOTE: data analysis module must use GPT-4.
        st.session_state["openai_model"] = "gpt-4"

    if "messages" not in st.session_state:
        st.session_state.messages = []

    for message in st.session_state.messages:
        with st.chat_message(message["role"]):
            st.markdown(message["content"])

    if prompt := st.chat_input("What is up?"):
        st.session_state.messages.append({"role": "user", "content": prompt})
        with st.chat_message("user"):
            st.markdown(prompt)

        with st.chat_message("assistant"):
            message_placeholder = st.empty()
            full_response = ""

            # 原始示例 https://docs.streamlit.io/knowledge-base/tutorials/build-conversational-apps
            # for response in openai.ChatCompletion.create(
            #     model=st.session_state["openai_model"],
            #     messages=[
            #         {"role": m["role"], "content": m["content"]}
            #         for m in st.session_state.messages
            #     ],
            #     stream=True,
            # ):
            #     full_response += response.choices[0].delta.get("content", "")
            # message_placeholder.markdown(full_response + "▌")

            async with CodeInterpreterSession() as session:
                # user_request = "对于文件中的'SepalLengthCm’数据给我一个'直方图'，提供图表，并给出分析结果"
                #! 可以用设定dpi=300来输出高质量的图表。（注：图的解析度dpi设定为300）
                # environ_settings = "【<默认要求> 如果我没有告诉你任何定制化的要求，那么请按照以下的默认要求来回答：1. 你需要用提问的语言来回答（即：如果我用中文提问，你就用中文来回答；我如果用英文提问吗，你就用英文来回答）。2. 如果要求你输出图表，那么图的解析度dpi需要设定为300。图尽量使用seaborn库。seaborn库的参数设定：sns.set(rc={'axes.facecolor':'#FFF9ED','figure.facecolor':'#FFF9ED'}, palette='deep')。】" ## seaborn中的palette参数可以设定图表的颜色,选项包括：deep, muted, pastel, bright, dark, colorblind，Spectral。更多参数可以参考：https://seaborn.pydata.org/generated/seaborn.color_palette.html。
                environ_settings = """【背景要求】如果我没有告诉你任何定制化的要求，那么请你按照以下的默认要求来回答：
                -------------------------------------------------------------------------
                1. 你需要用提问的语言来回答（如：中文提问你就用中文来回答，英文提问你就用英文来回答）。
                2. 如果要求你输出图表，那么图的解析度dpi需要设定为600。图尽量使用seaborn库。seaborn库的参数设定：sns.set(rc={'axes.facecolor':'#FFF9ED','figure.facecolor':'#FFF9ED'}, palette='deep', font='myfont')。
                3. matplotlib和seaborn图表上的字体需要设置为中文字体，如下：
                    plt.rcParams['font.sans-serif'] = ['myfont']
                    plt.rcParams['font.family'] = 'sans-serif'
                    mpl.rcParams['font.family'] = ['myfont']
                    sns.set(font='myfont')
                -------------------------------------------------------------------------
                """  # seaborn中的palette参数可以设定图表的颜色,选项包括：deep, muted, pastel, bright, dark, colorblind，Spectral。更多参数可以参考：https://seaborn.pydata.org/generated/seaborn.color_palette.html。

                # sns.set(font='myfont')

                user_request = environ_settings + "\n\n" + \
                    "你需要完成以下任务：\n\n" + prompt + f"注：文件位置在{uploaded_file_path}"
                # print('user_request: \n', user_request)

                # 加载上传的文件，主要路径在上面代码中。
                files = [File.from_path(str(uploaded_file_path))]

                with st.status('processing...', expanded=True, state='running') as status:
                    mpl.rcParams['font.family'] = 'myfont'
                    # sns.set(font='myfont')
                    # generate the response
                    response = await session.generate_response(
                        user_request, files=files
                    )

                    # output to the user
                    print("AI: ", response.content)
                    full_response = response.content
                    ### full_response = "this is full response"

                    # for file in response.files:
                    for i, file in enumerate(response.files):
                        # await file.asave(f"/Users/yunshi/Downloads/360Data/Data Center/Working-On Task/演讲与培训/2023ChatGPT/Coding/code_interpreter/output{i}.png") ##working.
                        # st.image(file.get_image() #! working.
                        # file.show_image()
                        # st.image(file.get_image(), width=500, output_format='png')
                        # * 注意这里的设定，可以提高图片的精细程度。
                        st.image(file.get_image(), width=None,
                                 output_format='PNG')

                    # message_placeholder.markdown(full_response + "▌") ## orignal code.
                # message_placeholder.markdown(full_response) ## orignal code.
                st.write(full_response)
                status.update(label='complete', state='complete')

            # TODO：看看是否缩进准确？ 放在这里好像可以。
            await session.astop()  # ! 确认需要关闭。

        st.session_state.messages.append(
            {"role": "assistant", "content": full_response})


if __name__ == "__main__":
    import asyncio
    # * 也可以用命令执行这个python文件。’streamlit run frontend/app.py‘
    asyncio.run(main())