Web Andromeda Instructions

Web Andromeda URL: http://cosmos.cs.vt.edu:8383/cosmos/andromeda.html
(Works best on Chrome & Safari)

Introduction

Andromeda is a interactive visual analytics system designed to aid users in the analysis of high-dimensional data. It implements the Weighted Multi-dimensional Scaling (WMDS) reduction algorithm, which reduces the high-dimensional data to two dimensions. WMDS uses a weighted Euclidean distance function, in which each attribute is weighted independently. Users can explore alternative projections by directly dragging the points in the “Observations” window or adjusting the attribute weight sliders in “Attributes” window.

Input Data

Andromeda can load data in CSV file format. The first row of the file should be a header containing the names of the attributes. Other rows contain the raw data for each observation and attribute. The first column must contain the observation names for labeling points in the projection. All other columns must contain quantitative data.

  • Default File (Animal Dataset)
    The dataset contains 49 animals and 72 attributes. Each attribute is measured on a 0 - 100 scale that describes how much it fits each animal. For example, a Grizzly Bear is very Furry, therefore its Furry value would be close to 100.

Buttons

  • Start Here
    • “Load Data”: You can create a new session by using files in the provided list, or click "Upload Data" button to upload your own CSV file.
    • “andromedaX”: Join an existing shared collaborative session named “andromedaX”.
  • Deselect All
    Deselect all highlighted data points. Change orange highlighed data points to original steel blue color.
  • Update Layout
    Compute a new projection based on orange highlighted data points, selected or moved by the user. Attribute weights are automatically updated to best match your desired new layout. You must select or move at least 2 points to enable this button, moving some closer and some further from each other to express desired changes in relative distances between points. Only the orange highlighted points are considered by the update computation, other points are ignored.
  • Reset
    Set the projection back to the initial configuration with all attributes weighted equally.
  • Information
    Link to this "web andromeda instructions" page.

Interactions

  • Observations
    You can hover, select and drag points in “Observation” window, then click “Update Layout” button to compute new projection and update attribute weights accordingly.
    • Hover
      Point will change to yellow when cursor hovers over a data point, and the raw data for hovered point can be viewed on attribute weight lines in “Attribute” window.
    • Select
      Single click to select and highlight point in orange color. Selected points are considered by the algorithm when updating a new layout.
    • Drag
      Move data points, moving some closer together and some further apart, to adequately express new relative distance relationships. Dragged data points will be highlighted in orange. And colored distance lines will show up when you are dragging points. Distance lines will change from green to dark blue to indicate the decreasing relative distances between the moving point and other highlighted points. On the other hand, distance lines will change from green to yellow to express the increasing relative distances between the moving point and other highlighted points.

    You must select or move at least 2 points to enable the “Update Layout” button. The algorithm only takes selected and moved points, which are highlighted in orange, into account when updating the layout. Because of this, you must select or drag points in relation to other selected or moved points. For example, if you want to move a single point closer to a cluster of points, all points in the cluster must be selected as well as the single point. All the points that are NOT highlighted in orange are ignored by the “Update Layout” button.

  • Attributes
    • By hovering, selecting and dragging data points in “Observations” window, you can view the raw data of those data points on attribute weight lines. The hovered point indicator is in yellow color, and selected and moved indicators are in orange color.
    • Hovering on an attribute weight handle (gray circle handles on the right end of attribute weight lines) will change the size of all data points in the “Observation” window based on their attribute values. For example, Sheep is more furry than Elephant, therefore Sheep have a larger data point size than Elephant when hovering on the Furry attribute handle.
    • Dragging attribute weights (gray circle handles on right end of attribute weight lines) will automatically update the projection to reflect the new set of weights in the weighted distance function. Dragging handle to the right will increase the attribute weight, while dragging handle to the left will decrease the weight. The attribute weights are automatically arranged in descending order.

Contact: Ming Wang (mingw [at] vt [dot] edu)