General ideas for my project

[Vinnie Gelb]

Hello!

I am currently studying mechanical engeneering in germany and will begin my bachelor thesis soon.

Also I am pretty new to python and machine learning in general, but have great interest in this topic. I would like to ask a question of the general idea of my project. Sorry in advance if this is a silly one:

It's all about injection molding.
I have got a dataset of maschine settings and the corresponding parameters of the finished injection molded part. Meaning i have 6 parameters to set and about 12 measurements like size and weigth of the finished part.

From my understanding of genetic programming I either need to have a function which corresponds the machine settings to the final product (which I don't have) or train a neural network with hidden layers to find out how the settings influence the measurements.

I hope you get my point...

To my question:
How do I implement hidden layers with the deap library? And how do I do this without the need of a function connecting the settings to the  final product?

Thank you in advance and have a nice weekend!
Vinnie

[EBo]

Vinnie,

I have heard people say that there are no stupid/silly questions unless
you know the answer and are bating people for an hyper critical
response. Your question is nothing like that, and I think you will find
people here quite helpful.

To help us answer your questions, can you give us some idea of what you
are working on. Even if you do not know the structure and/or
characteristics of your "fitness function", it is likely that you and
others can reason out using simple principles of physics what it is
likely to be. Everything is orders of approximation. I do not know if
you have ever heard the joke about the physicist starting reasoning a
problem with the words "let us first examine the case of a spherical
cow..." (see: Think Like a Physicist
https://scienceblogs.com/principles/2013/07/09/think-like-a-physicist),
but there are lots of things that spherical cow can tell us. Once you
have some simple models of the world, you can go back and remove one
assumption after another until you get better and better results. One
place where I have used DEAP and SCOOP is to parameterize the Coupled
Routing and Excess STorage model (CREST) hydrological model. What you
might not understand is that no matter how good or bad that model might
be, it is ONLY a model of the world and not the real thing. In that
case the fitness function was given to me. I ran the model and compared
a rivers outflow to the model results. So, if we have some idea of what
you are trying to model/calibrate/optimize, we might be able to point
you to existing mathematical models/equations. That said, your advisor,
or one of the other university professors, would be a good resource to
talk to about this. I would also suggest speaking to your research
librarian to help find resources.

I have never worked on neural networks, and cannot speak to them, so I
will leave others to comment there.

I wish you the best of luck!

EBo --