Graph Editor

The Graph Editor provides a visual interface for building data pipelines.

Overview

The graph shows your data flow as connected nodes:

  • Layer nodes (rectangles) - Data containers

  • Transformation nodes (rounded) - Operations between layers

  • Projection nodes (circles) - Dimension reduction for viewing

Nodes are connected by edges showing the data flow direction.

Node Types

Source Layers (Green)

Your original data. These can be:

  • Uploaded files (CSV, NPY, NPZ)

  • Generated synthetic data

  • sklearn datasets

Source layers show column configuration in the Config Panel.

Derived Layers (Blue)

Created when you add a transformation. These inherit the points from the source layer but have different vector values.

Transformation Nodes (Purple/Orange)

Operations that connect layers:

  • Scaling (purple) - Multiply vectors by a factor

  • Rotation (orange) - Rotate in a 2D plane

  • Affine - Linear transformation with translation

  • Linear - Matrix multiplication

Projection Nodes (Blue/Purple)

Dimension reduction for visualization:

  • PCA (blue) - Principal Component Analysis

  • t-SNE (purple) - t-distributed Stochastic Neighbor Embedding

Interacting with Nodes

Selecting

Click on any node to select it. The Config Panel on the right shows:

  • Node name (click to edit)

  • Node type and properties

  • Type-specific controls

  • Actions (Add View, Add Transformation, etc.)

Panning and Zooming

  • Pan: Click and drag the background

  • Zoom: Scroll wheel

Node Layout

Nodes are automatically positioned but can be manually arranged by dragging.

Config Panel

When a node is selected, the Config Panel shows:

For Layers

  • Name (editable)

  • Type (Source/Derived)

  • Point count

  • Dimensionality

  • Column configuration (source layers only)

  • Existing views

  • “Add View” section

  • “Add Transformation” section (if no outgoing transformation)

For Transformations

  • Name (editable)

  • Type selector

  • Type-specific parameters (sliders for scale, angle, etc.)

  • Invertibility status

For Projections

  • Name (editable)

  • Type and dimensions

  • Random seed (for reproducibility)

  • “Show View” button - opens the View Editor

Building a Pipeline

Example: Scaled PCA View

  1. Load data (creates source layer)

  2. Click the layer node

  3. Add a Scaling transformation (creates derived layer)

  4. Click the new derived layer

  5. Add a PCA view

Now you have: Source → Scaling → Derived Layer → PCA View

Example: Compare Original and Transformed

  1. Load data

  2. Add a PCA view to the source layer

  3. Add a Transformation (scaling, rotation)

  4. Add a PCA view to the derived layer

  5. Switch to Viewports mode to compare both views

Tips

  • One transformation per layer: Each layer can only have one outgoing transformation, but can have multiple projections

  • Chain transformations: To apply multiple transformations, chain them: Source → T1 → Layer → T2 → Layer

  • Real-time updates: Changing transformation parameters immediately updates derived layers and projections