Feature multilayer perceptron and data-driven fluid model

[EvertBunschoten]

Proposed Changes

Addition of multi-layer perceptron class which can be used to evaluate trained multi-layer perceptrons in processes such as thermodynamic state evaluation in data-driven fluid models.

Related Work

PR Checklist

Put an X by all that apply. You can fill this out after submitting the PR. If you have any questions, don't hesitate to ask! We want to help. These are a guide for you to know what the reviewers will be looking for in your contribution.

In the Tutorials repository, branch feature_multilayer_perceptron, one can find under "common/multilayer_perceptron" some python scripts which demonstrate how to translate a model trained through Tensorflow to an input file compatible with the CLookUp_ANN class. The README.md file in that tutorial folder also lists some important information regarding the functionality of the CLookUp_ANN class, datadriven fluid model, and MLP input file format.

Shall I leave those there or should these be uploaded elsewhere?

• I am submitting my contribution to the develop branch.
• My contribution generates no new compiler warnings (try with --warnlevel=3 when using meson).
• My contribution is commented and consistent with SU2 style (https://su2code.github.io/docs_v7/Style-Guide/).
• I have added a test case that demonstrates my contribution, if necessary.
• I have updated appropriate documentation (Tutorials, Docs Page, config_template.cpp), if necessary.

[lgtm-com]

This pull request introduces 6 alerts when merging 9370536 into 45214cd - view on LGTM.com

new alerts:

• 4 for Resource not released in destructor
• 1 for Non-virtual destructor in base class
• 1 for Constant return type on member

[pcarruscag]

Do you want to introduce this first as a standalone library, and then start using it for fluid models?

[EvertBunschoten]

@pcarruscag Yes I want to introduce the multilayer perceptron as an option for users and developers for things like data-driven fluid models. Do you think it would be good to add a template CFluidModel child class demonstrating how the MLP class can be used to create data-driven fluid models (apart from writing a tutorial of course)?

[pcarruscag]

Sounds good. Initially it would be enough to have some unit tests that would already show how to setup the network, together with documentation / example of the file format. Applications can come after.

Some initial comments:
Please move the files to toolboxes/ and ideally use a namespace for the new classes.
Start the class names with C as we do, e.g. CIOMap.

Then, how large are the models you've used so far? and how important is performance to this feature? (Just so I know how much to comment on that)

[EvertBunschoten]

Of course I'll provide an example of an MLP input file, as well as a python script I wrote that translates an MLP trained through TensorFlow to such an input file.

Very well, I'll move the files from the numerics folder to the common folder and change names accordingly.

The models I used so far had between 5 and 50 perceptrons and up to 15 layers. Performance is quite important, as evaluation of MLP's is generally more computationally expensive than for example lookup tables. The larger the MLP architecture, the more costly evaluations get of course. Any improvement to the computation speed will therefore be welcome. In terms of memory, the MLP's don't seem to be an issue so far.

[lgtm-com]

This pull request introduces 3 alerts when merging 47b456c into 88c8392 - view on LGTM.com

new alerts:

• 3 for Resource not released in destructor

[lgtm-com]

This pull request introduces 3 alerts when merging 88f0012 into d32ccec - view on LGTM.com

new alerts:

• 3 for Resource not released in destructor

[lgtm-com]

This pull request introduces 3 alerts when merging 04813b1 into 124795b - view on LGTM.com

new alerts:

• 3 for Resource not released in destructor

[lgtm-com]

This pull request introduces 3 alerts when merging 112876d into 7132d99 - view on LGTM.com

new alerts:

• 3 for Resource not released in destructor

---

Heads-up: LGTM.com's PR analysis will be disabled on the 5th of December, and LGTM.com will be shut down ⏻ completely on the 16th of December 2022. Please enable GitHub code scanning, which uses the same CodeQL engine ⚙️ that powers LGTM.com. For more information, please check out our post on the GitHub blog.

[lgtm-com]

This pull request introduces 3 alerts when merging e8a3a93 into 7132d99 - view on LGTM.com

new alerts:

• 3 for Resource not released in destructor

---

Heads-up: LGTM.com's PR analysis will be disabled on the 5th of December, and LGTM.com will be shut down ⏻ completely on the 16th of December 2022. Please enable GitHub code scanning, which uses the same CodeQL engine ⚙️ that powers LGTM.com. For more information, please check out our post on the GitHub blog.

[bigfooted]

@EvertBunschoten I suggest to first finish and close this one before moving on to the hydrogen flamelets PR

[pcarruscag]

I'll ignore the request for review the same way you ignored my questions 🤷

[EvertBunschoten]

I'll ignore the request for review the same way you ignored my questions shrug

Could you clarify your questions? I sent you a message on Slack.

[pcarruscag]

Evert thought he was replying but the messages were pending and never made it online.