flare.cnn
cnn-1D-feats
(cnn-1D-feats model input-dim kernel-specs)
module that builds max-pool conv features.
The output is concattened height of all kernel-specs which has the max value of the conv-1D across the sliding windows
conv1D-builder
(conv1D-builder model input-dim kernel-spec)
builds a function which takes a sequence of input nodes and returns the sequence of convolutional features over each window
Input: model: PModel
to add kernel params to inputs: seq of n
input nodes kernel-spec: width, height, stride (defaults to 1) of kernel
Outputs: Lazy sequence of size (partition width stride inputs)
nodes each with dimension height
from kernel heights