Projection

Overview

The Projection Tool within the Mantis platform equips users to visualize, transform, and reconfigure high-dimensional embeddings into interpretable 2D or 3D spaces. It enables intuitive exploration, comparative embedding overlays, and active retraining workflows.

Purpose

• Translate complex embedding spaces into explorable visual projections.
• Provide dimensionality reduction for spatial clarity and structure.
• Overlay, compare, and contrast multiple projection techniques.
• Anchor real-time experimentation in active learning workflows.

Key Features

• Project Data: Choose dimensionality reduction techniques like t-SNE, UMAP, or PCA for embedding.
• Overlay Embeddings: Compare multiple projection layers side-by-side or stacked.
• Custom Parameters: Modify perplexity, distance metrics, number of neighbors, etc.
• Live Updates: Changes to projection settings reflect instantly in your map interfaces.
• Projection Recall: Save, name, and reuse projection configurations across sessions.

How to Use

• Project Data

  1. Open the Machine Learning panel.
  2. Navigate to the Projection Tool tab.
  3. Select your method (UMAP, t-SNE, PCA).
  4. Set the projection parameters.
  5. Click Project to render new embeddings.

• Overlay Embeddings

  1. Toggle "Overlay Projections" in the same tab.
  2. Load previously saved projections or auto-generated ones.
  3. Analyze layered structure or transitions over time.

• Save or Reuse Projections

  1. After projecting, click Save Projections.
  2. Enter a name and optional tags.
  3. Retrieve saved projections from Saved Projections.

Tips

• Use UMAP for faster runtime and broad structural preservation.
• Try t-SNE when investigating local clusters and fine-grained similarities.