Thank you for your answer. But this solution doesn’t seem to work… my shapes stop matching.
Maybe I wasn’t clear on my question. Let me try and clarify it:
Consider my X to have these columns (features/predictors):
beta = ['-windowstyle', 'hidden', '-executionpolicy', 'bypass', 'powershellscript', '-nologo', 'remove-update', 'invoke-ssidexfil', '-payload']
I have given them all a prior distribution of ~N(0,1), then did score = beta@X and then fit this to a sigmoid curve to get the probability my sample belongs to class 1.
However, consider that I have gotten domain information that ['remove-update', 'invoke-ssidexfil', '-payload'] are features that increase the probability of my sample to belong to class 1; then I would like to give them a ~N(1,0.5) distribution, to increase how much they push my end probability closer to 1.
Might be relevant to know that the beta.shape = (721,) and I have around 400 samples.
EDIT:
I checked that and indeed the concatenate just includes these 3 betas in the end, which causes the order of the columns of X to not match the orders of beta anymore.
I then fixed this by enforcing the last 3 columns to be ['remove-update', 'invoke-ssidexfil', '-payload'] so that beta and X are correct, I tested it with both betas~N(0,1) and it seems that this fixed the issue…
…however, the outcome is a little bit different than model_1 but I would expect them to be the same. Any ideas why?
- Does the order of my features in
Xorbetainfluence the result? I would imagine that it doesn’t… - Every time I train my model, even if I do not change anything, it might yield results that are a little bit different because of how MCMC works?