search

EthanJamesLew / AutoKoopman

AutoKoopman - automated Koopman operator methods for data-driven dynamical systems analysis and control.
GNU General Public License v3.0
53 stars 10 forks source link

Bug: Older AutoKoopman is More Accurate #70

Closed EthanJamesLew closed 5 months ago

EthanJamesLew commented 1 year ago

I have been trying to run the Koopman falsification code on a different machine and have noticed quite different results. The main branch is giving different results than the version I have (which I installed straight after the fix you provided regarding the time step). I am attaching two figures highlighting this, with the result after one iteration learning the Vanderpool model. The old version learns a very accurate representation whereas the current main does not. I haven’t been able to locate the source of error yet so figured it was best to flag it with you.

image

image

EthanJamesLew commented 1 year ago

@Abdu-Hekal

EthanJamesLew commented 1 year ago

@Abdu-Hekal to provide a timeline: I will look at this tomorrow evening. I am quite swamped in other things atm. Sorry for the delay.

Abdu-Hekal commented 1 year ago

No worries, working on the older version for now!

EthanJamesLew commented 1 year ago

CC @KochdumperNiklas

EthanJamesLew commented 1 year ago

@Abdu-Hekal can you help me reproduce your issue? I have attempted to test autokoopman for the case you show, notebook here: https://colab.research.google.com/drive/18UPHoEZ0kTxknfUcwqNJarFcr_cOBtb8?usp=sharing

Abdu-Hekal commented 1 year ago

@EthanJamesLew I have updated the notebook to reproduce the issue. This issue seems to happen when there is a lesser number of trajectories (I have changed training data to 4 trajectories).

EthanJamesLew commented 1 year ago

@Abdu-Hekal Awesome! I see this download (1)

However, the new training data got deleted when starting a new colab runtime. May you please email me the data?

Abdu-Hekal commented 1 year ago

Sure thing! Are these the ones you are after?

From: Ethan Lew @.> Date: Wednesday, 7 June 2023 at 9:00 PM To: EthanJamesLew/AutoKoopman @.> Cc: Abdelrahman Waleed Elsayed Aly Hekal (PGR) @.>, Mention @.> Subject: Re: [EthanJamesLew/AutoKoopman] Bug: Older AutoKoopman is More Accurate (Issue #70)

⚠ External sender. Take care when opening links or attachments. Do not provide your login details.

@Abdu-Hekalhttps://github.com/Abdu-Hekal Awesome! I see this [Image removed by sender. download (1)]https://user-images.githubusercontent.com/15683738/244183475-dd285be2-8958-40da-8046-c46708c9f71e.png

However, the new training data got deleted when starting a new colab runtime. May you please email me the data?

— Reply to this email directly, view it on GitHubhttps://github.com/EthanJamesLew/AutoKoopman/issues/70#issuecomment-1581427936, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AVZPXM24WI5VCXJ2MQFS7NTXKDMWBANCNFSM6AAAAAAYJOE2UU. You are receiving this because you were mentioned.Message ID: @.***>

EthanJamesLew commented 5 months ago

We resolved this and the falsification now uses the latest autokoopman