Datasets
Organize and version your training data.
Creating a Dataset
Navigate to Deep Learning Platform → Datasets → Click Create
Basic Information
Dataset Name* (Required)
Enter a descriptive name for the dataset
Example:
imagenet-1k,coco-2017,custom-dataset
Version* (Required)
Semantic version (major.minor.patch)
Default:
1.0.0Helper text: "Semantic version (major.minor.patch)"
Description* (Required)
Detailed description of dataset contents and purpose
Dataset Configuration
Dataset Type* (Required)
Select data type from dropdown:
Tabular
Image
Text
Audio
Video
Others
Default:
tabular
Task Type* (Required)
Select task type:
Classification
Regression
Detection
Segmentation
Others
Default:
classification
Data Format* (Required)
Select file format:
CSV
JSON
Parquet
TFRecord
Others
Default:
csv
Status* (Required)
Current dataset status:
Uploading
Processing
Ready
Failed
Default:
uploading
Metadata
Tags (Optional)
Comma-separated tags for organization
Example:
computer-vision, training, augmented
License (Required)
Select license:
MIT
CC BY 4.0
CC0
Apache 2.0
Proprietary
Default:
MIT
Public Access (Checkbox)
Make dataset accessible to all organization members
Actions
Cancel: Discard and close
Create Dataset: Submit and create the dataset
Example Configuration
Viewing Dataset Details
To view detailed information about a dataset:
Navigate to Deep Learning Platform → Datasets
Click on a dataset from the list
View comprehensive details in the modal dialog
Details Panel Sections:
Basic Information:
Dataset Name: e.g., "Time Series Sales Data"
Version: Semantic version (e.g., 1.1.0)
Description: Full description of the dataset
Dataset Configuration:
Dataset Type: Time Series, Tabular, Image, Text, Audio, Video
Task Type: Forecasting, Classification, Regression, etc.
Data Format: CSV, JSON, Parquet, TFRecord, etc.
Status: Processing, Uploading, Ready, Failed
Metadata:
Tags: Comma-separated tags (e.g., "time-series,forecasting,sales,multivariate,business")
License: Proprietary, MIT, CC BY 4.0, etc.
Public Access: Checkbox for visibility
Editing a Dataset
To update dataset information:
Open dataset details page
Click Edit button (or three-dot menu → Edit)
Modify editable fields in the Edit Dataset modal
Click Update Dataset to save changes
[!NOTE] The Edit form is identical to the View form, but with editable fields and an "Update Dataset" button.
Editable Fields:
✅ Description
✅ Tags
✅ License
✅ Public Access
❌ Dataset Name (cannot edit)
❌ Version (create new version instead)
❌ Data Type (cannot edit)
❌ Data Format (cannot edit)
Downloading Dataset
To download dataset files:
Open dataset details
Click Download button
Select download format:
Original format
Compressed archive
Specific splits (train/val/test)
Download will start
Download Options:
Full dataset
Specific splits only
Sample subset
Metadata only
Creating a New Version
To create a new version of a dataset:
Open dataset details
Click New Version button
Enter new version number
Upload updated data
Document changes in description
Click Create
When to Version:
Added new samples
Fixed data errors
Changed preprocessing
Updated annotations
Removed corrupted files
Deleting a Dataset
To remove a dataset:
Navigate to dataset details
Click Delete button
Confirm deletion
[!WARNING] You cannot delete a dataset that is being used by running experiments. Stop experiments first.
Before Deleting:
Check for experiments using this dataset
Download data if needed
Update documentation
Consider archiving instead
Dataset Versioning
When to Create New Version:
Added new samples
Fixed annotation errors
Changed preprocessing
Updated splits
Removed corrupted data
Best Practices
Organization:
Use consistent naming conventions
Document data collection process
Include data cards/datasheets
Provide sample data for preview
Quality Control:
Validate data integrity
Check for label errors
Monitor class balance
Document known issues
Next Steps
Use dataset in Experiments
Track data lineage
Monitor usage in Analytics
Last updated