5803025018/Tugas_Kecerdasan_Buatan_10_5803025018
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Delete tugas week 10..py

Browse Source
This commit is contained in:
5803025018
2026-11-01 22:38:45 -05:00
parent a57ed8005d
commit 6b3ca4ac5c
1 changed files with 0 additions and 91 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tugas week 10..py
-91
View File
@@ -1,91 +0,0 @@
import os
import chromadb
from chromadb.utils import embedding_functions
from pypdf import PdfReader
from langchain_text_splitters import RecursiveCharacterTextSplitter
# 1. Persiapan Dokumen
pdf_path = "dokumen_akademik.pdf" # Ganti dengan nama file PDF yang disiapkan
print(f"Membaca file: {pdf_path}")
reader = PdfReader('Buku Pedoman Akademik.pdf')
raw_text = ""
for page in reader.pages:
# Menggabungkan seluruh teks dari PDF
raw_text += page.extract_text() + "\n"
# 2. Setup Embedding Function
# Menggunakan model multilingual lokal agar performa pencarian untuk Bahasa Indonesia lebih akurat
print("Menyiapkan model embedding (ini akan mengunduh model pada run pertama)...")
emb_fn = embedding_functions.SentenceTransformerEmbeddingFunction(
model_name="paraphrase-multilingual-MiniLM-L12-v2"
)
# 3. Inisialisasi ChromaDB Client
# PersistentClient akan membuat folder 'chroma_db_data' untuk menyimpan vektor secara lokal
client = chromadb.PersistentClient(path="./chroma_db_data")
# Eksperimen 3 ukuran chunk sesuai instruksi modul
chunk_sizes = [500, 1000, 1500]
for size in chunk_sizes:
print(f"\n--- Memproses ukuran chunk: {size} karakter ---")
collection_name = f"rag_koleksi_chunk_{size}"
# Hapus collection jika sebelumnya sudah ada saat di-run ulang
try:
client.delete_collection(name=collection_name)
except Exception:
pass
# Buat collection baru di vector DB
collection = client.create_collection(
name=collection_name,
embedding_function=emb_fn
)
# 4. Setup Text Splitter
# Overlap disetel 10% dari ukuran chunk untuk mencegah konteks terpotong mendadak
text_splitter = RecursiveCharacterTextSplitter(
chunk_size=size,
chunk_overlap=int(size * 0.1),
length_function=len,
)
# Potong teks menjadi list of chunks
chunks = text_splitter.split_text(raw_text)
print(f"Total chunk yang dihasilkan: {len(chunks)} chunks")
# 5. Persiapkan ID dan Metadata
ids = [f"id_chunk_{size}_{i}" for i in range(len(chunks))]
metadatas = [{"source": pdf_path, "chunk_size": size, "chunk_index": i} for i in range(len(chunks))]
# 6. Masukkan Vektor, Teks, dan Metadata ke Vector Database
collection.add(
documents=chunks,
metadatas=metadatas,
ids=ids
)
print(f"Data chunk {size} berhasil di-index ke ChromaDB!")
# ==========================================
# CONTOH PENCARIAN (QUERY)
# ==========================================
print("\n--- Menguji Pencarian (Query Retrieval) ---")
query_text = "Apa syarat untuk mengajukan cuti akademik?" # Sesuaikan dengan isi PDF yang di-load
# Kita uji pencarian di collection yang ukuran chunk-nya 1000
test_collection = client.get_collection(name="rag_koleksi_chunk_1000", embedding_function=emb_fn)
# Retrieve Top-K (K=2)
results = test_collection.query(
query_texts=[query_text],
n_results=2
)
print(f"Pertanyaan: '{query_text}'\n")
# Hasil dibungkus di dalam list of lists oleh chromadb, maka kita akses index [0]
for i, (doc, dist) in enumerate(zip(results['documents'][0], results['distances'][0])):
print(f"--- Top {i+1} (Jarak Vektor: {dist:.4f}) ---")
print(f"{doc}\n")