引導式測驗 · 中等 等級
ML 資料的 Parquet 與資料欄格式測驗
了解 Parquet 的列塊、行組、壓縮和修剪,然後根據實際的 ML 工作負載測試效能。
問題 1 的 8
How is data organized in a Parquet file?
預覽每個問題8 問題
當您參加測驗時,就會出現答案和解釋。
- How is data organized in a Parquet file?
- Only as comma-separated text lines
- As one unstructured binary blob
- As one database row per file
- Into row groups containing column chunks, which contain pages
- Why can columnar storage help a query that reads only a few fields?
- Every row is stored as a separate file
- A reader may fetch selected columns without reading every column’s data
- It automatically removes all null values
- It guarantees a faster model
- What role can Parquet statistics or page indexes play?
- They provide a full database index for every query
- They guarantee zero I/O
- They may help compatible readers skip pages or row groups that cannot match a filter
- They store model weights
- What does a row group contain?
- A column chunk for each column in that logical row partition
- Only metadata with no data
- One row only
- A separate model checkpoint
- When might Parquet be less suitable than another storage design?
- For any machine-learning workload
- When data contain numeric values
- When data are larger than a spreadsheet
- For workloads dominated by tiny random row lookups or frequent single-row updates
- Why benchmark on the target engine and storage system?
- All Parquet readers behave identically
- Compression never affects performance
- Feature support and performance depend on reader, layout, and workload
- Benchmarks are unrelated to data format
- What can improve statistics-based pruning?
- Randomizing every row across all files
- Organizing data so common filter columns have useful value ranges
- Removing column metadata
- Increasing CSV quoting
- Does Parquet alone provide table transactions and dataset version history?
- No; those require surrounding table-management or catalog features
- Yes, every Parquet file has ACID transactions
- Yes, it stores every prior dataset version forever
- No, because Parquet cannot store schemas