GRU Cell 2D: Gate Traces & Unit Heatmap

A Gated Recurrent Unit (GRU) is the leaner cousin of the LSTM — the same idea of gating what a recurrent network remembers, done with two gates and no separate cell state. This 2D simulator runs a real 8-unit GRU cell over a live input sequence you choose (sine, square, impulse, or noise), but renders it in a way that is genuinely native to two dimensions rather than a repainted 3D scene: a scrolling oscilloscope shows the input and gate averages evolving over time, while a flat heatmap grid shows every one of the 8 hidden units individually at the current step. Two bias sliders let you push the reset and update gates toward always-open or always-closed so you can watch, directly, how a GRU decides what to forget and what to keep — and the gate identities used here (a closed reset gate makes the candidate provably independent of history; a closed update gate provably freezes the hidden state) are checked with a standalone numerical test, not just asserted.