Skip to content
Eric Riebling edited this page Dec 5, 2018 · 1 revision

k-cross.sh is main script found in /home/engaclew/trainers on Bridges

sbatch k-cross.sh (can override params with -P) multiple pcas live in 1 folder future: maybe use

gpu-shared slower tha -p GPU-AI maybe increase hours

eval-test.sh takes .mat file which is ccross-fold validation thingy model = folder name

in gru200-adam-lr1e-03-const-seed1 net.epoch247.all.pt is the model .pt is model that goes into vm

can't evaluate just k-fold outputs on what it was trained on

result is a confusion matrix (and others)

collapse.py maps classes of corpora to FAT MOT CHI OTH e.g.

-- .

Clone this wiki locally