search

AndreWeiner / ml-cfd-lecture

Lecture material for machine learning applied to computational fluid mechanics
GNU General Public License v3.0
330 stars 116 forks source link

Check/test of exercise 3/4 part I #23

Closed AndreWeiner closed 1 year ago

AndreWeiner commented 1 year ago

Hi @JanisGeise,

I pushed a revised version of exercise 3 (next week's exercise). I didn't have time to test the instructions thoroughly. Please let me know if further revision is needed. The revision of the second part and the lecture notebook will have to wait until Monday.

Thanks! Andre

JanisGeise commented 1 year ago

Hi @AndreWeiner,

the first part of the exercise works without any problems. In my opinion, all instructions are clear and understandable.

Regards Janis

AndreWeiner commented 1 year ago

Great, thanks for testing! I update this issue once there are updates on the lecture/exercise. Best, Andre

AndreWeiner commented 1 year ago

Hi @JanisGeise, a small update: the revised lecture notebook is now online. The explanations towards the end of the notebook are sparse due to time constraints. If there is something really essential to write some more lines about, please let me know. Thanks! Andre

JanisGeise commented 1 year ago

Hi @AndreWeiner,

in my opinion (if it's possible in the remaining time) it would be helpful to add one or two more sentences when explaining overfitting (prior section "Training with batches of data").

Regards, Janis

AndreWeiner commented 1 year ago

Hi @JanisGeise, I completed and pushed exercise 5, the second part of the ML intro exercise. I'll upload a quick&dirty solution to your cloud storage in case you get stuck. Best, Andre

JanisGeise commented 1 year ago

Hi @AndreWeiner,

I finished exercise 5, everything worked fine. I have one suggestion though:

In my opinion, it would be helpful to add a note to the "Direct learning approach" part (as tip / advice) to start with a rather "simple" model for the baseline model, because it may be confusing which of the presented models is best to use as a baseline model since the lecture notebook covers a wide range of methods up to quite complex techniques.

Unrelated to that, I added a function for the hyperparameter tuning to your solution script and uploaded it into the cloud, in case you need it.

Regards, Janis

AndreWeiner commented 1 year ago

Hi @JanisGeise, thanks for the script - I'll show the function to perform the parameter study if requested and time allows. I included a note to start with a simple network and a simple training routine for the baseline model. Best, Andre