"""退款资料检索与 LCEL 入门示例。

按文章第 5 节安装依赖后运行。首次使用需要下载本地嵌入模型，
运行时会在当前目录写入 refund-demo-notes.txt。生成回答使用固定回复的模拟模型。
"""

from pathlib import Path
from langchain_core.documents import Document
from langchain_text_splitters import RecursiveCharacterTextSplitter

path = Path("refund-demo-notes.txt")
path.write_text(
    "购买后 30 天内，可以在订单页面提交退款申请。\n\n"
    "退款审核通过后，款项将在 3 个工作日内原路退回。\n\n"
    "员工请假需要在办公系统提交申请，由直属主管审批。",
    encoding="utf-8",
)

text = path.read_text(encoding="utf-8")
documents = [Document(
    page_content=text,
    metadata={"source": path.name},
)]

splitter = RecursiveCharacterTextSplitter(
    chunk_size=40,
    chunk_overlap=0,
)
chunks = splitter.split_documents(documents)

print("片段数：", len(chunks))
for chunk in chunks:
    print(chunk.page_content)

from langchain_huggingface import HuggingFaceEmbeddings
from langchain_core.vectorstores import InMemoryVectorStore

embeddings = HuggingFaceEmbeddings(
    model_name="BAAI/bge-small-zh-v1.5",
    model_kwargs={"device": "cpu"},
    # 主动设置：把生成的向量归一化
    encode_kwargs={"normalize_embeddings": True},
)

# 使用上面的嵌入组件；这个内存库固定用余弦相似度搜索
store = InMemoryVectorStore(embedding=embeddings)
store.add_documents(chunks)

vector = embeddings.embed_query("钱怎么退回来？")
print("向量维度：", len(vector))  # 向量维度：512

retriever = store.as_retriever(search_kwargs={"k": 2})
question = "钱怎么退回来？"
docs = retriever.invoke(question)

for doc in docs:
    print(doc.page_content)

for doc, score in store.similarity_search_with_score(question, k=3):
    print(f"{score:.4f}", doc.page_content)

parts = []
for doc in docs:
    parts.append(doc.page_content)
context = "\n\n".join(parts)

from langchain_core.prompts import ChatPromptTemplate
from langchain_core.language_models.fake_chat_models import FakeListChatModel
from langchain_core.output_parsers import StrOutputParser

prompt = ChatPromptTemplate.from_template(
    "请根据资料回答问题，资料不足时说明无法确认。\n"
    "资料：\n{context}\n\n问题：{question}"
)
model = FakeListChatModel(responses=[
    "可以在订单页面申请退款；审核通过后，款项将在 3 个工作日内原路退回。"
])
parser = StrOutputParser()

inputs = {"context": context, "question": question}
prompt_value = prompt.invoke(inputs)
message = model.invoke(prompt_value)
answer = parser.invoke(message)
print(answer)

chain = prompt | model | parser
answer = chain.invoke(inputs)
print(answer)
