Skip to content

A tool that performs 3D embedding of data and provides interactive visualization.

License

Notifications You must be signed in to change notification settings

sonjageorgievska/Embed-Dive

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

15 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Please cite the software if you are using it in your scientific publication.

DOI

Please also cite the paper on LargeVis as stated here.

Embed-Dive

Embed-Dive is a pipeline consisting of LargeVis and DiVE. Please visit the respective repositories for more information, individual licencing, latest versions etc.

The purpose of this repository is to provide a release of the pipeline to (for the moment) Linux users that want to explore their data, provided that the input is a similarities graph.

Screenshot

##Usage - for Linux users ##

Download the compiled version of Embed-Dive_linux from [here] (https://github.com/sonjageorgievska/Embed-Dive/releases) in the same folder where you have downloaded Embed_Dive.

For quick try and test:

./Embed_Dive_linux -input test/sim.txt -output test/coord.txt -outdim 3 -samples 1

At the end you should be able to see your browser opening the visualization.

Required parameteres:

  • -input: Input file of similarities (see the test folder for input format).
  • -output: Output file of low-dimensional representations.

Besides the two parameters, other optional parameters include:

  • -threads: Number of threads. Default is 8. Recommended: cores x 2.

  • -outdim: The lower dimensionality LargesVis learns for visualization (2 or 3). Default is 2.

  • -samples: Number of edge samples for graph layout (in millions). Default is set to data size / 100 (million).

  • -prop: Number of times for neighbor propagations in the state of K-NNG construction, usually less than 3. Default is 3.

  • -alpha: Initial learning rate. Default is 1.0.

  • -trees: Number of random-projection trees used for constructing K-NNG. 50 is sufficient for most cases unless you are dealing with very large datasets (e.g. data size over 5 million), and less trees are suitable for smaller datasets. Default is set according to the data size.

  • -neg: Number of negative samples used for negative sampling. Default is 5.

  • -neigh: Number of neighbors (K) in K-NNG, which is usually set as three times of perplexity. Default is 150.

  • -gamma: The weights assigned to negative edges. Default is 7.

  • -perp: The perplexity used for deciding edge weights in K-NNG. Default is 50.

  • -metaData: file containing meta information about data. Format: [id] [metadata]. Format of metadata: "first_line" "second_line" "third_line" (number of lines is not limited). Example line of metadata: 35 "A dog" "Age:2" "Color brown".

  • -dir: base directory to store output file

  • -divedir: directory where DiVE resides

  • -np: A json file containing list of properties names. Ex: ["Height", "Weight", "Place of birth"].

  • -json: Name of json file which is input to DiVE. It can be uploaded at anytime later from DiVE.

Best practices:

  • Do not use the -samples parameter; let the embedding use the default value. Use -samples 1 only if you want quick results.
  • Use -outdim 2 unless the visualization results are very different than with -outdim 3.
  • The similarities graph does not need to contain duplicate edges (does not need to be bi-directional)
  • Use the -json parameter so that you can easily load the coordinates in DiVE at any later stage. Computing the coordinates with LargeVis takes a lot of time, it only needs to be done once.
  • If you forgot to include metadata before computing coordinates, you can combine the coordinates with metadata later as described here DiVE (""From output of LargeVis to input of DiVE")

Please send any suggestions for improvement to the author.

Licence

The software is released under the Creative Commons Attribution-NoDerivatives licence. Contact the author if you would like a version with an Apache licence