This is a layout algorithm based on a force directed system (a more sophisticated version of Fruchterman-Reingold), optimized for dealing with large networks and highly configurable. There are many forces acting on the nodes: repulsion on every pair of nodes depending on their degree, attraction on linked nodes, central gravity, outer attraction and more.
|Iterations*||int > 0||500||Number of iterations of the algorithm.|
|Outbound Attraction Distribution||Bool||False||Forces all nodes to move away from the center.|
|Node radius||float > 0||1.0||Minimum node radius, used to prevent overlap. It can be used to expand dense node clusters without increasing much the total network size.|
|Link lengths||float > 0||0||Equilibrium length of the links. If it is ‘0’, this spring force is not considered.|
|Scaling ratio||float > 0||1.0||Global expansion factor.|
|Barnes Hut Theta||float > 0||1.0||Acceleration method for the N-body problem. Reduce its value to have a more physically accurate system, but computationally more challenging, and increase it to approximate repulsion forces and increase computation speed.|
|Is in lin/log Mode||Bool||True||Whether use the logarithmic of the node distance to calculate attraction forces.|
|Link weight||text||None||Link property that represents the intensity of the attraction between connected nodes. Must have positive values|
|Link weight influence||float > 0||1.0||Increases the link weight effect on the attraction force.|
|Is Strong Gravity Mode||Bool||False||Activates a much stronger attractive central force to keep all nodes close to the center. It is recommended to use it with disconnected networks.|
|Gravity||float > 0||10.0||Intensity of the gravity force, which attracts the nodes to the center of coordinates proportionally to their mass.|
|Jitter tolerance||float > 0||1.1||Controls the node oscillations around their equilibrium point. Reduce it to damp the oscillations and obtain smoother node clusters.|
|Mass range||float > 0||200.0||Global multiplicative constant for the repulsion force. Use it to expand the network without any performance variation.|
|2D||Bool||False||Whether to use a 2D or 3D layout.|
|Seed||int||0||Random seed. If equals zero, it is chosen by the system.|
|Reset Positions||Bool||False||Whether to delete any previously calculated positions and start a new fresh layout.|
* Required Field
Since there are many input parameters, this layout requires some expertise and trial-error tests to obtain nice visualizations.
Here we show two examples of Force Atlas results applied on two social networks. We can observe how the nodes group together in clusters around some important nodes. Node color correspond to the Louvain community detection algorithm.