One Hot Encoder — Machine Learning/Preprocessing

One Hot Encoder

Machine Learning / Preprocessing

Turns a class index into an indicator column: the selected element is 1, every other element is 0.

The categorical front end of any model whose inputs are not all continuous – a mode selector, a gear, a fault code – and the exact inverse of Argmax Decision. The index arrives as an ordinary signal and is rounded to the nearest whole number, so a value of 2.0 and one of 1.9997 select the same class.

Ports

• k – the class index, a scalar [1,1]. Interpreted relative to Index Base.
• Output – y, a column [K,1] where K is Number Of Classes. Exactly one element is 1 whenever the index is in range.

Parameters

• Number Of Classes – K, the output height; a whole number ≥ 1. Rounded to the nearest whole number if you give it a fraction.
• Index Base – what the first class is called:
  • Zero-based (PyTorch, numpy) – index 0 selects the first element. The default.
  • One-based (MATLAB) – index 1 selects the first element.
• Out Of Range – what an index outside the K classes does:
  • All Zeros – the output is entirely zero, which is sklearn's handle_unknown='ignore'. The default, and the one that lets a downstream block notice that nothing was selected.
  • Clamp To Range – below the first class selects the first, above the last selects the last. Choose this only when every out-of-range value genuinely belongs to the nearest class.
  This is a real decision, not a corner case: an unmapped category or a glitched sensor produces an out-of-range index on a live system, and the two settings send it to opposite places.
• Sampling Time (s) – zero or less inherits the solver's rate; a positive value runs the block at that period.

Code export

All ten targets: Python, MATLAB, Java, Rust, C, C++, VHDL, Verilog, SystemVerilog and PLC Structured Text. K, the base and the out-of-range rule are all structural and baked in; there is no tunable parameter.

No backend converts the index to an integer. Each element is emitted as a half-open comparison – (k − base) − i in −0.5, +0.5) – which is the same function as floor(x + 0.5) == i without needing an integer type. Every integer cast in these ten languages truncates toward zero, which would send a negative index to class 0 instead of out of range.

The three HDL targets are ordinary Q16.16 fixed point and are offered for synthesis: one subtraction and a comparison per class.

Simulink bridge

None. Simulink has no one-hot block: Multiport Switch routes signals by an index rather than producing an indicator, and the Deep Learning blocks take a trained network object. The bridge reports the block rather than dropping it silently, and it has no parity testbench, which is the documented consequence of Support::None rather than a gap.

Notes

• Algebraic and stateless: the output depends only on the current input, so the block cannot break an algebraic loop.
• Nonlinear and discontinuous, and deliberately carries no state space.
• The rounding boundary is a discontinuity, and the HDL targets sit one quantum away from the reference. They compare Q16.16 values (one quantum ≈ 1.5e-5) where the in-app run compares doubles, so an index within that distance of a ±0.5 boundary can round the other way and select a different class. Feed this block a genuinely integral index – which is what a category is – and the question never arises; feed it a continuous quantity and it is inherent, not a codegen defect.