Skip to main content

Zip folder layout

Organize zip archives so Lake unstructured ingest can label files automatically. Processed datasets include filename and label columns.

Before you start
  • Ingest via Lake → Add data → Unstructured (guide)
  • Top-level folder names become class labels

How labeling works​

  • Filename — name of each file inside the zip
  • Label — taken from the top-level folder that contains the file
  • Folder names — become class labels for all files in that folder

Image classification example​

Fire detection with two classes:

ForestFireImages.zip/
├── Fire/
│   ├── fire1.png
│   ├── fire2.jpg
│   └── fire3.png
└── NoFire/
    ├── forest1.png
    ├── forest2.jpg
    └── trees1.png

Text classification example​

Sentiment analysis:

SentimentData.zip/
├── Positive/
│   ├── review1.txt
│   └── review2.txt
└── Negative/
    ├── complaint1.txt
    └── complaint2.txt

Supported file types​

Images​

  • JPG / JPEG
  • PNG
  • GIF

Text​

  • TXT (plain text)

Best practices​

Organization​

  • Use one folder per class at the zip root
  • Avoid deep nesting — keep structure flat
  • Use clear, consistent file names

Data quality​

  • Remove corrupted or duplicate files
  • Balance classes when possible (similar counts per folder)
  • Use readable images and clean text

Next steps​

  1. Create your zip following the layout above
  2. Run unstructured ingest
  3. Curate silver or publish gold for ML
Pro tip

Well-organized zips produce cleaner bronze tables and faster paths to gold for ML.