Table 3.

Symbols used to describe FIDDLE’s implementation

Symbol	Shape	Description
$N$	–	The number of examples.
$L$	–	The number of time bins, calculated as $\lfloor T/dt\rfloor$.
$\widehat{d}$, $\widehat{D}$	–	The number of input variables that are time-invariant/time-dependent after the pre-filter step.
$\stackrel{\sim }{d}$, $\stackrel{\sim }{D}$	–	The dimensionalities of time-invariant / time-dependent features after the transform step and before the post-filter step.
$d$, $D$	–	The final dimensionalities of time-invariant / time-dependent features.
$\widehat{\mathbf{S}}$	$N\times \widehat{d}$	Data tables containing values of raw time-invariant/time-dependent values after the pre-filter step.
$\widehat{\mathcal{X}}$	$N\times L\times \widehat{D}$
$\stackrel{\sim }{\mathbf{S}}$	$N\times \stackrel{\sim }{d}$	Tensors containing the time-invariant/time-dependent features for all $N$ examples, after the transform step and before the post-filter step.
$\stackrel{\sim }{\mathcal{X}}$	$N\times L\times \stackrel{\sim }{D}$
$\mathbf{S}$	$N\times d$	Tensors containing the final time-invariant/time-dependent features for all $N$ examples.
$\mathcal{X}$	$N\times L\times D$

Note: FIDDLE: Flexible Data-Driven Pipeline.