In our project each datapoint is the result of an experiment done in the lab (some of these take 90 days) so obviously we don't have thousands of points. After some reading and reserch we are aware of the importance when you have few data of being extrimely careful with overfitting, identifying outliers, giving priority to simple models, validate with CV leave one out...
The questions are, which techniques are the best for handling the lack of data? Can we really trust leave one out CV or is there a better way to estimate the error of our model?
The lab is also interested in us creating confidence intervals (or other measurements to give reliability for their experiments) for the data they have provided us. Could you give us some references (code libraries and theory) where we could start reading?
Few data, how to squeeze it to the fullest?
Hello all!
In our project each datapoint is the result of an experiment done in the lab (some of these take 90 days) so obviously we don't have thousands of points. After some reading and reserch we are aware of the importance when you have few data of being extrimely careful with overfitting, identifying outliers, giving priority to simple models, validate with CV leave one out...
The questions are, which techniques are the best for handling the lack of data? Can we really trust leave one out CV or is there a better way to estimate the error of our model?
The lab is also interested in us creating confidence intervals (or other measurements to give reliability for their experiments) for the data they have provided us. Could you give us some references (code libraries and theory) where we could start reading?
Thank you in advance :)
1
(Answered in Discord)
https://discord.com/channels/755347620175675452/755347620704288790/776869824034373634
Add comment