Streamlit 佈署機器學習網頁應用

# 連線到Google雲端硬碟
from google.colab import drive
drive.mount('/content/drive', force_remount=True)

# 建立並切換到Apps目錄
%mkdir '/content/drive/My Drive/Applied-NLP-in-the-Enterprise/apps'
%cd '/content/drive/My Drive/Applied-NLP-in-the-Enterprise/apps'

%%writefile ner_app_agnews.py

# 載入必要的函式庫
import spacy
import streamlit as st
import spacy_streamlit
import numpy as np
import pandas as pd
import random

# 設定標題
st.title(':star: AG News資料集')

# 定義讀取資料的函式
@st.cache
def read_data(file):
    read_path = '/content/drive/My Drive/Applied-NLP-in-the-Enterprise'
    data = pd.read_csv(read_path+file)
    return data

# 讀取資料
data = read_data('/data/ag_dataset/prepared/train_prepared.csv')

# 定義選擇類別的函式
@st.cache
def select_category(data, category_option):
    return data.loc[data.class_name==category_option]

# 設定側邊欄控制元件
st.sidebar.header("引數")
display_selections = st.sidebar.multiselect(
    "您想顯示哪些資料？",
    ["完整資料","單篇文章"], None)
category_option = st.sidebar.radio(
    '您想探索哪個類別？',
    data.class_name.unique())
article_option = st.sidebar.number_input(
    '您想探索哪篇文章？',
    data.loc[data.class_name==category_option].index.min(),
    data.loc[data.class_name==category_option].index.max(),
    data.loc[data.class_name==category_option].index.min())

# 設定資料的顯示行為
if "完整資料" in display_selections:
    st.header("完整資料")
    st.write(select_category(data, category_option).loc[:, ["title", "description"]])

if "單篇文章" in display_selections:
    st.header("單篇文章")
    st.subheader("標題")
    st.write(data.loc[article_option, "title"])
    st.subheader("描述")
    st.write(data.loc[article_option, "description"])

# 設定文字
default_text = data.loc[article_option, "description"]

# 顯示命名實體辨識結果
st.header("命名實體辨識結果")
base, custom = st.beta_columns(2)

# 基本spaCy模型
with base:
    base_model = spacy_streamlit.load_model("en_core_web_lg")
    doc_base = base_model(default_text)
    ner_labels = ["ORG", "PERSON", "GPE"]
    show_table = True
    title = "基本SpaCy模型"
    sidebar_title = "基本SpaCy模型"
    spacy_streamlit.visualize_ner(doc_base, labels=ner_labels, show_table=show_table, title=title, sidebar_title=sidebar_title)

# 自定義spaCy模型
with custom:
    custom_ner_model = spacy_streamlit.load_model('/content/drive/My Drive/Applied-NLP-in-the-Enterprise/models/ag_dataset/ner-base-V3/model-best')
    doc_custom = custom_ner_model(default_text)
    ner_labels = ["ORG", "PERSON", "GPE", "TICKER"]
    show_table = True
    title = "自定義SpaCy模型"
    sidebar_title = "自定義SpaCy模型"
    spacy_streamlit.visualize_ner(doc_custom, labels=ner_labels, show_table=show_table, title=title, sidebar_title=sidebar_title)

# 安裝必要的函式庫
!pip install -U spacy[cuda100]
!pip install -U spacy-lookups-data
!pip install cupy-cuda100==7.3.0

# 下載預訓練語言模型（核心模型）
!python -m spacy download en_core_web_lg

# 安裝spacy-streamlit
!pip install spacy-streamlit

# 安裝 Streamlit
!pip install streamlit
# 安裝 ngrok
!pip install pyngrok

# 連線到 Google Drive
from google.colab import drive
drive.mount('/content/drive', force_remount=True)

# 定義佈署函式
def deploy(path):
    # 終止已存在的 ngrok 隧道
    try:
        ngrok.kill()
        print("所有 ngrok 工作已終止。")
    except:
        print("沒有 ngrok 工作需要終止。")
    
    # 設定應用程式位置
    app_location = path
    
    # 匯入 ngrok
    from pyngrok import ngrok
    
    # 建立一個隧道到 Streamlit 的 80 連線埠
    public_url = ngrok.connect(port='80')
    print(public_url)
    
    # 執行 Streamlit 應用程式
    !streamlit run $app_location --server.port 80 >/dev/null

# 佈署應用程式
deploy('/content/drive/"My Drive"/Applied-NLP-in-the-Enterprise/apps/ner_app_agnews.py')

%%writefile ner_app_custom.py
# 載入函式庫
import spacy
import streamlit as st
import spacy_streamlit
import numpy as np
import pandas as pd
import random

# 設定標題
st.title(':star: Custom NER')
st.header("Custom Text")
default_text = st.text_area("輸入要分析的文字。", height=250)

# 顯示 NER 結果
st.header("NER Results")
base, custom = st.beta_columns(2)

# 基本 spaCy 模型
with base:
    base_model = spacy_streamlit.load_model("en_core_web_lg")
    doc_base = base_model(default_text)
    ner_labels = ["ORG","PERSON","GPE"]
    show_table = True
    title = "Base SpaCy Model"
    sidebar_title = "Base SpaCy Model"
    spacy_streamlit.visualize_ner(doc_base, labels=ner_labels, show_table=show_table, title=title, sidebar_title=sidebar_title)

# 自定義 spaCy 模型
with custom:
    custom_ner_model = spacy_streamlit.load_model('/content/drive/My Drive/Applied-NLP-in-the-Enterprise/models/ag_dataset/ner-base-V3/model-best')
    doc_custom = custom_ner_model(default_text)
    ner_labels = ["ORG","PERSON","GPE","TICKER"]
    show_table = True
    title = "Custom SpaCy Model"
    sidebar_title = "Custom SpaCy Model"
    spacy_streamlit.visualize_ner(doc_custom, labels=ner_labels, show_table=show_table, title=title, sidebar_title=sidebar_title)