# Datasets

Organize and version your training data.

![Datasets View](/files/Gt4nM53UipUUAhPNHEL1)

## Creating a Dataset

Navigate to **Deep Learning Platform** → **Datasets** → Click **Create**

![Create Dataset Form](/files/LONaMkad0GcuoZlIVAqh)

### 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: `1.0.0`
* Helper 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

```yaml
Dataset Name: imagenet-subset-2024
Version: 1.0.0
Description: ImageNet subset with 100 classes for quick experimentation
Dataset Type: Tabular
Task Type: Classification
Data Format: CSV
Status: Uploading
Tags: computer-vision, subset, training
License: MIT
Public Access: ✓
```

## Viewing Dataset Details

To view detailed information about a dataset:

1. Navigate to **Deep Learning Platform** → **Datasets**
2. Click on a dataset from the list
3. View comprehensive details in the modal dialog

![View Dataset Details](/files/imNTrVTP5D0nKZQV8q61)

**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:

1. Open dataset details page
2. Click **Edit** button (or three-dot menu → Edit)
3. Modify editable fields in the Edit Dataset modal

![Edit Dataset Form](/files/IlMO5yX1CfrVWPcUldVb)

4. 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:

1. Open dataset details
2. Click **Download** button
3. Select download format:
   * Original format
   * Compressed archive
   * Specific splits (train/val/test)
4. 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:

1. Open dataset details
2. Click **New Version** button
3. Enter new version number
4. Upload updated data
5. Document changes in description
6. 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:

1. Navigate to dataset details
2. Click **Delete** button
3. 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](/kaisar-network/kaisar-ai-ops/deep-learning-platform/experiments.md)
* Track data lineage
* Monitor usage in [Analytics](/kaisar-network/kaisar-ai-ops/deep-learning-platform/analytics.md)


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.kaisar.io/kaisar-network/kaisar-ai-ops/deep-learning-platform/datasets.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.