Embed

Embed performs Takens delay embedding: one channel x(t) becomes points (x(t), x(t−τ), x(t−2τ)…) in as many dimensions as you ask for. The geometry of the resulting cloud is the dynamics — a periodic signal traces a loop, a chaotic one a strange shape, noise a formless ball.

Lag at 0 picks τ automatically, where the autocorrelation first crosses zero — the standard heuristic. Dimension sets how many delayed copies are stacked.

Two outputs: Points gives the first three dimensions normalised to Grafit units, ready for Points Visualize, Plexus, particles or Plot's Scatter and XY; the Signal carries every dimension as a channel, for further analysis such as Recurrence.

• See the attractor — A sine becomes a circle, a heartbeat a distinctive loop, noise a blob.
• Distinguish signals that look alike — Two traces with the same spectrum can have obviously different attractors.
• Feed Recurrence — The embedded Signal is the proper input for a recurrence plot of a single channel.
• Generative form — The attractor of a live audio or EEG channel as a point cloud you draw with.
• Detect a change in dynamics — The shape changes before the statistics do.

• Let Lag choose itself first; override it only when you know the process's timescale.
• Dimension 3 is what you can see. Higher dimensions are for Recurrence and analysis, not for the eye.
• Filter first: embedding raw noisy data mostly shows you the noise.
• Max Points caps the cloud — lower it when feeding a visual consumer that does not need thousands.
• Lag Time reports τ in seconds, which is often itself an interesting measurement.