The graph is the computation of the model. Each card is a node: its
name on top, what it is below. Arrows carry values from one
node to the next, and are labelled with the shape of what they carry.
Nodes
Arrows
Shapes
A shape is written (seq…, tw, dim…), the batch
left out. seq are sequence axes, such as the steps of a rollout,
and dyn marks a length set by the data; tw
is the window, how many samples are read at once; dim are the features
of each sample.
Views
Standard: every sub-model, a model called inside another or a
Loop, is one node. Double-click it to open its body on a page of its own.
Flattened: every sub-model is spliced in, its body inside a shaded
region, and the recurrence of a Loop drawn as a feedback arrow.
Exploring
Click a node to see what it is, its shape and its connections; click an arrow for what it carries.
From the panel, highlight everything a node depends on or feeds.
Click an entry of the legend to highlight every node of that kind.
Scroll to zoom, drag the background to pan, drag a node to move it.
/ search · Enter next match · F fit · Esc clear · ? this help