Length — transformers/Length.py

Non-zero indicator transformer. Converts each element to a binary indicator (1 if non-zero, 0 if zero), takes the row-wise max, and appends the transform salt.

mask_i = (SD_i != 0) ? 1 : 0 eta = max(mask_i, dim=1) SLT_i = Concat(eta, TS_i)

This effectively outputs 1.0 if any element of the record is non-zero and 0.0 if the record is all zeros (i.e., a default record).

Constructor

python Length(size=0, length_transform_salt=0)

Parameter	Type	Default	Description
size	int	0	Expected output shape.
length_transform_salt	int	0	Dimensionality of TS_i.

forward(a, salt)

Parameter	Type	Description
a	torch.Tensor	Salted raw value SD_i, shape (1, record_dim).
salt	torch.Tensor	Transform salt TS_i, shape (1, salt_dim).

Returns torch.Tensor of shape (1, 1 + salt_dim) — Concat(eta, salt).

Note

Used as one of the two transformers in the Logistic Accuracy module (LogisticAccuracy) to count the number of non-default records (i.e. the sample size N). The root value of the resulting SMT equals N in its first element.

---

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