bpmweel
commented
3 years ago Here are the timings we measured when we taught it:
Day 1
- 9:30 Started episode 1
- 10:35 Break 1 : check installation. Finished exercise 1 and explanation about deep learning, not yet deep learning workflow Couldn't resume after 15 min because fixing setups took more time. Breakout Exercise Deep learning workflow needs a bit more than 10 min (15 min seems OK)
- 11:10 Finish episode 1
- 11:37 Break 2: just finished pairplotting exercise
- 12:50 Finish: finished Exercise: Create the neural network (silent docing)
Day 2
- 9:15 start recap + Q&A yesterday
- 9:35 start choosing a cost function
- 9:51 start training history exercise
- 10:00 finish training history exercise
- 10:22 Confusion matrix exercise
- 10:31 Confusion matrix exercise stops, start break
- 10:53 Stop break, Start weather prediction
- 11:05 Start data inspection exercise weather prediction
- 11:?? Stop exercise
- 11:30 Start exercise train/test/val set
- 11:40 Stop exercise
- 11:47 Start break
- 12:05 Start ‘build a neural network’ exercise
- 12:22 Finish exercise
- 13:00 Finished training the model and inspecting the evaluation on train and test set
Day 3
• 9:19 start recap yesterday • 9:31 start exercise create a similar scatterplot for a reasonable baseline • 9:45 ?? Stop exercise • 10:07 start exercise plotting validation history • 10:12?? Stop exercise • 10:20 Start exercise less parameters in breakout rooms • 10:30 Stop exercise less parameters • 10:50 Start explaining early stopping • 11:05 Start exercise batchnorm • 11:20?? Stop exercise batchnorm • 11:30 Start exercise dropout • 11:48 Stop exercise dropout • BREAK • 12:07: Discuss dropout exercise • 12:12 saving and loading model (start) • 12:20??: start exercise on how to further improve the model • 12:35??: stop exercise • 12:50 start survey
Update the episode timings based on what we measured during the june workshop.