How can I extract information from continuous data and use it as predictors in a bayesian model?
Thanks for any help!
That’s quite a wide-open question, do you mean time-series data?
There’s a handful of example notebooks here that might help: examples_notebooks — PyMC3 3.10.0 documentation see the sections for Time Series and (if you’re feeling brave) Gaussian Processes
Hello jonsedar!
I work with spectroscopy, so my continuous data are matrices that contain energy variation (intensities) as a function of wavelength.
They are not time series but the examples that you mentioned can be applied to my data type?
Thanks for your answer!
Ah okay, interesting - a million years ago I spent a few months working with an SEM, while my colleague across the hall used a huge XPS machine. Both analyzing silicon on glass. Anyhow.
Are you hoping to do some pattern recognition on the spectra? Quantify uncertainty somehow?
1 Like
Yes, I would like to make a calibration model and predict the concentration of organic compounds.
If I were to make this model with an artificial neural network, I could apply Principal Component Analysis to this matrix of spectra and use the eigenvalues obtained as input to the model and the concentration of the organic compound as the output.
I don’t know if the eigenvalues have a Normal distribution and if they could be used as predictors of a Bayesian model. They can?
Thank you!
So to paraphrase (if I may), you want to decompose a signal into a set of features that you can then dump into a (linear | non-linear) model and regress onto a continuous target feature?
Sounds like regular old generalised linear regression and is absolutely possible. You might find some food for thought in the examples Notebooks https://docs.pymc.io, and some general statistics reading including:
2 Likes