decision boundary
epoch 0 · train – · test –
epoch 0 | train –
● A ● B ○ –what to noticewhat to look for
The boundary bends where the data forces it and stays straight where it doesn't. Eight units a layer is comfortable for this spiral; drop layers or units and watch it give up on one arm first - capacity runs out from the outside in.
instruments
loss
log scale– train –-- test –
gradient magnitude by layer
Gradients shrink as they travel back – the products-of-slopes fact from Maths of Learning ch3, live.