CS390-NIP Lab 2: Convolutional Neural Networks

---

Run python lab2.py -h to see usage instructions Example: python lab2.py -a tf_conv -d mnist_f -e 20

Viewing Learning Curves via Tensorboard

Run tensorboard --logdir logs/fit to startup local tensorboard, reachable at localhost:6006 once started

• Recorded logs are stored in /logs/old
• Logs are grouped in "rounds"
  • A round consists of saved logs for all five datasets, run on the same architecture of network
• To view a log, move the desired folder from /logs/old/round/ to /logs/fit/, then refresh tensorboard
  • May need to create /logs/fit/ folder
  • To view more than one dataset at a time, move all desired logs into /logs/fit/

Viewing Saved Models

Run python lab2.py -l path-to-saved-model

• Saved models are stored in /models
• Models are grouped in "rounds"
  • A round consists of saved models for all five datasets, run on the same architecture of network
• Each saved model has a meta.txt file which consists of a quick summary of hyperparameters and accuracy

Example: python lab2.py -l ./models/round3/tf_conv-cifar_100_f-2020-10-05-22.10.25