BinCount — transformers/BinCount.py

Histogram bin-count transformer. Assigns each scalar data record to a histogram bin defined by a set of edges, producing a one-hot bin-membership vector, then appends the transform salt. Used for the KS test.

bin_mask = one_hot_bin_assignment(SD_i[0], edges) SLT_i = Concat(bin_mask.flatten(), TS_i)

The scalar feature a[:, 0] is linearly rescaled from the source range [0, 31] to the edge range before binning.

Constructor

python BinCount(size=0, length_transform_salt=0, edges=None)

Parameter	Type	Default	Description
size	int	0	Expected output shape.
length_transform_salt	int	0	Dimensionality of TS_i.
edges	torch.Tensor	None	1-D tensor of bin boundary values. n_bins = len(edges) - 1.

forward(a, salt, edges)

Parameter	Type	Description
a	torch.Tensor	Salted raw value SD_i, shape (1, record_dim). Only a[:, 0] is used as the scalar feature.
salt	torch.Tensor	Transform salt TS_i.
edges	torch.Tensor	Bin boundaries.

Returns torch.Tensor of shape (1, n_bins + salt_dim).

Internal steps:

1. Extract scalar feature: x = a[:, 0].
2. Rescale from [0, 31] to [edges[0], edges[-1]].
3. Compute bin_mask: a float indicator per bin — 1.0 if the scaled value falls in [lower, upper).
4. Zero out bins for rows where the entire input is zero (default records).
5. Flatten and concatenate with salt.

Warning

Only the first column of a is used. All other columns (user salt) are ignored by this transform.

---

See index.md for the shared interface inherited by all transformers.