max_pool1d_op

→TensorImplPtr

int max_pool1d_op(const int & x, int KL, int stride_l, int pad_l)

One-dimensional max pooling over a batch of sequences.

Emits the maximum of each sliding window of length KL along the last axis of the input.

xTensorImplPtr

Input of shape (B, C, L).

KLint

Window length.

stride_lint= None

Stride. 0 is a sentinel meaning "use KL" (non-overlapping windows). Default: 0.

pad_lint= None

Zero-padding added to both ends. Default: 0.

TensorImplPtr

Output of shape (B, C, L_out) where $L_\text{out} = \lfloor (L + 2 p - KL)/s + 1 \rfloor$ .