Ccc / app.py
william1324's picture
Update app.py
d328e9b verified
Raw History Blame Contribute Delete
7.22 kB
import tempfile
import gradio as gr
import pandas as pd
import spaces
from apriori_core import (
apriori,
association_rules,
default_text,
format_items,
parse_csv,
parse_text,
validate_threshold,
verify_mlxtend,
)
FREQUENT_COLS = [
"商品組合",
"出現次數",
"Support(支持度)"
]
RULE_COLS = [
"前件(Antecedent)",
"後件(Consequent)",
"Support",
"Confidence",
"Lift"
]
def save_csv(frame, prefix):
with tempfile.NamedTemporaryFile(
mode="w",
encoding="utf-8-sig",
newline="",
suffix=".csv",
prefix=prefix,
delete=False
) as file:
frame.to_csv(file, index=False)
return file.name
@spaces.GPU(duration=15)
def analyze_ui(
source,
transaction_text,
uploaded_csv,
minimum_support,
minimum_confidence
):
try:
minimum_support = validate_threshold(
minimum_support, "最小支持度"
)
minimum_confidence = validate_threshold(
minimum_confidence, "最小信賴度"
)
if source == "文字輸入":
data = parse_text(transaction_text)
elif source == "上傳 CSV":
data = parse_csv(uploaded_csv)
else:
raise ValueError("請選擇有效的資料來源")
counts = apriori(data, minimum_support)
rules = association_rules(
counts,
len(data),
minimum_confidence
)
verification_message = verify_mlxtend(
data,
counts,
rules,
minimum_support,
minimum_confidence
)
except (ValueError, AssertionError, OSError, ImportError) as error:
raise gr.Error(str(error)) from error
frequent_rows = []
for itemset, count in sorted(
counts.items(),
key=lambda x: (
len(x[0]),
tuple(sorted(x[0]))
)
):
frequent_rows.append({
"商品組合": format_items(itemset),
"出現次數": count,
"Support(支持度)": f"{count / len(data):.2%}"
})
rule_rows = []
for rule in rules:
rule_rows.append({
"前件(Antecedent)": format_items(rule["前件"]),
"後件(Consequent)": format_items(rule["後件"]),
"Support": f"{rule['Support']:.2%}",
"Confidence": f"{rule['Confidence']:.2%}",
"Lift": f"{rule['Lift']:.4f}"
})
frequent_df = pd.DataFrame(
frequent_rows,
columns=FREQUENT_COLS
)
rules_df = pd.DataFrame(
rule_rows,
columns=RULE_COLS
)
unique_items = len(set().union(*data))
summary = (
"### 分析完成\n"
f"交易筆數:**{len(data)}**\n\n"
f"商品種類:**{unique_items}**\n\n"
f"頻繁項目集:**{len(counts)} 組**\n\n"
f"關聯規則:**{len(rules)} 條**\n\n"
f"最小支持度:**{minimum_support:.0%}**\n\n"
f"最小信賴度:**{minimum_confidence:.0%}**\n\n"
f"### mlxtend 交叉驗證\n{verification_message}"
)
if not rules:
summary += (
"\n\n目前沒有符合條件的關聯規則,"
"請降低支持度或信賴度。"
)
return (
summary,
frequent_df,
rules_df,
save_csv(frequent_df, "frequent_"),
save_csv(rules_df, "rules_")
)
def reset_defaults():
return (
"文字輸入",
default_text(),
None,
0.3,
0.7
)
with gr.Blocks(
title="Apriori 關聯規則分析系統"
) as demo:
gr.Markdown(
"""
# Apriori 關聯規則分析系統
Python 自行實作 Apriori 演算法
功能:
- 自訂購買紀錄
- CSV 資料上傳
- 支持度與信賴度設定
- 頻繁項目集分析
- 關聯規則分析
- mlxtend 交叉驗證
- CSV 結果下載
"""
)
with gr.Row():
with gr.Column(scale=3):
source = gr.Radio(
choices=["文字輸入", "上傳 CSV"],
value="文字輸入",
label="資料來源"
)
transactions_input = gr.Textbox(
value=default_text(),
lines=15,
label="購買紀錄",
info="每行一筆交易,商品以逗號分隔,至少10筆"
)
csv_input = gr.File(
label="上傳 CSV 檔案",
file_types=[".csv"],
type="filepath"
)
with gr.Column(scale=2):
min_support = gr.Slider(
minimum=0.05,
maximum=1.0,
value=0.3,
step=0.05,
label="Minimum Support(最小支持度)"
)
min_confidence = gr.Slider(
minimum=0.05,
maximum=1.0,
value=0.7,
step=0.05,
label="Minimum Confidence(最小信賴度)"
)
run_button = gr.Button(
"開始分析",
variant="primary"
)
reset_button = gr.Button(
"恢復預設資料"
)
gr.Markdown(
"""
### 指標說明
**Support(支持度)**
商品組合同時出現在交易中的比例。
**Confidence(信賴度)**
購買前件商品後,也購買後件商品的比例。
**Lift(提升度)**
衡量商品間的關聯程度。
- Lift > 1:正向關聯
- Lift = 1:獨立
- Lift < 1:負向關聯
"""
)
summary = gr.Markdown(
"請點選開始分析"
)
with gr.Tab("頻繁項目集"):
frequent_table = gr.Dataframe(
label="Frequent Itemsets",
interactive=False
)
frequent_download = gr.File(
label="下載頻繁項目集 CSV",
interactive=False
)
with gr.Tab("關聯規則"):
rule_table = gr.Dataframe(
label="Association Rules",
interactive=False
)
rules_download = gr.File(
label="下載關聯規則 CSV",
interactive=False
)
run_button.click(
fn=analyze_ui,
inputs=[
source,
transactions_input,
csv_input,
min_support,
min_confidence
],
outputs=[
summary,
frequent_table,
rule_table,
frequent_download,
rules_download
]
)
reset_button.click(
fn=reset_defaults,
outputs=[
source,
transactions_input,
csv_input,
min_support,
min_confidence
]
)
if __name__ == "__main__":
demo.launch()