owned this note   changed 2 years ago  
Linked with GitHub

CUQIpy software training, 7-8 Dec 2023

DTU, Lyngby Campus, Denmark - Building 303A, auditorium 49

Entrance to Building 303A from Matematiktorvet.

Tentative programme

Thursday 7 December

Time Activity
09:00 Welcome (Per Christian)
09:10 Introduction to UQ for inverse problems with CUQIpy (Jakob)
10:10 Coffee break
10:30 Installation and getting started (Nicolai)
10:50 Exercise 1: Distributions (Amal)
12:00 Lunch (voucher provided)
13:00 Exercise 2: Forward models (Charlie)
14:45 Coffee break
15:15 Exercise 3: Bayesian inverse problems (Nicolai)
16:45 Wrap-up of the day (Jakob)
17:00 Plans for day 2 and mini project ideas (Jakob)
17:30 Free pizza and drinks event

Friday 8 December

Time Activity
09:00 CUQIpy open-ended training & mini projects
10:30 Coffee break
10:45 CUQIpy open-ended training & mini projects
12:00 Lunch (voucher provided)
13:00 CUQIpy open-ended training & mini projects
14:30 Wrap-up and evaluation
15:00 End of Day 2

Resources

Plugins

DTU Jupyter notebook servers

If not installing CUQIpy on own computer, you can use one of the two learnmore servers (login details provided at event):

When logged in to the learnmore servers you should see a "launcher" tab with some icons to start new notebooks with different python packages installed. You can use the following ones:

  • CUQIpy
  • CIL (has CUQIpy-CIL plugin)
  • CUQIpy-PyTorch

In the launcher you also see an icon "terminal", click that to start a linux terminal. In the you type/copy-paste:

git clone https://github.com/CUQI-DTU/CUQIpy-demos.git

This will clone/copy the CUQIpy demos to your directory, then you can navigate in the left menu into the training folder and start the training notebooks. You need to change the "kernel" in the top right corner to e.g. the "CUQIpy" one, in order for the notebook to have access to the python packages installed in there such as CUQIpy.

Alternatively you can use Google colab:

If using Google colab, add a link to the CUQIpy demos repository:
https://github.com/CUQI-DTU/CUQIpy-demos
and run command

!pip install cuqipy

Ideas for mini projects

List project topics here and please add your name to any project you are interested in working on:

  1. Gibbs sampling
    • Name
  2. UQ in tomography with CUQIpy-CIL (Negin)
  3. PDE-based inverse problem with CUQIpy-FEniCS
  4. Hamiltonean Monte-Carlo sampling with CUQIpy-pyTorch
  5. …

Feedback

  • We greatly appreciate it if you can take few moments to let us know about your CUQIpy training experience in this post-training feedback form: https://tinyurl.com/TellCUQI

Acknowledgments

  • This work was supported by The Villum Foundation (grant no. 25893).
  • Thanks to all CUQI project members for valuable input and contributions to CUQIpy!
  • And thank you for participating in the CUQIpy training! Please let us know what you think about the software and the training using the the feedback link just above.

Software issues, problems, comments:

Please let us know about any and all problems, suggestions, etc. about CUQIpy that you may have:

Did you discover any bugs or strange behaviour?

  • The function sample_adapt(x) of the sampler MH crashes for small x (maybe for x<10?). In this function, x is multiplied with 0.1, floor() called and then used as a denominator

Feature wishlist

Documentation issues

Other

Select a repo