Hey Happy People,
I’m completely new to huggingface and so far I’m super excited and fascinated. I dont have too much experience with modeling, but am really interested in learning more.
I would love to hear some tips and advices to my problem from you!
So I’m currently working on a project (in University as a student) in which we are producing an enzyme in a fermentation process. We can judge the quality of the product by a measuring method which gives a spectrum with different peaks.
There will be 8 different batches (with different cultivation parameters). Every batch is represented as a time series dataset with about 7-8 features and about 400 single datapoints. To every batch I will have 7 spectra which will allow me to evaluate the quality of the product. The 7 spectra a linked to different timestamps in the dataset.
I want to build a model which is capable of predicting the outcome of a batch, which means predicting the spectrum measurement, so the quality of the product. Based only on the starting parameters.
There is one big question arising for me.
How do I get to train the model properly? I need a target value for every datapoint, how can I do that? There is the possibility to interpolate the missing data points, but I think this might not be it. Also there is something called Gaussian processes with which you can do that, but I still need to get to know about this stuff better. Do you have any Ideas on how to deal with such a dataset?
What would your general approaches to this situation be and do you have any suggestions to what kind of model would be suitable for such a task?
(Also people have used hybrid forms of mechanistical- and ML-models to predict biological processes. If someone has useful information on how to integrate something like this, I would be happy to hear that.)