Is it feasible to define the model by utilizing both training and test data when partitioning the data, and subsequently sample posteriors based solely on the training data, followed by running a posterior predictive on the testing data? Alternatively, are we compelled to explicitly retrieve the pos…

You can definitely run the whole mapping over everything, why not? You just then need to split the map into train and test parts, and include data objects for the map in your model. I see a lot of places where the example you posted can be simplified. First, I organized the data into a DataFarme, a…

I think it would be useful if you could provide a simple example where you’re struggling with set_data, as the question is quite abstract. In general, everything your model needs to get parameters to the right spot is data, so you need to either put it all into mutable data containers, or re-structu…

updated the question with a small example, the example is very simplistic however. I think based on your answer however that we could restructure the data in some sort of way but lets say i have some term that modelles a carry-over effect between datapoints, it can get really really complex, there i…

i love this community… amazing response. 5/5

I edited some small mistakes, but glad to help. I understand your real model is much more complex, but hopefully this illustrates the principle. For each row of your data, make an integer array that maps it to the coefficient dimension you care about (in this case, which day-knot goes to which row)…

I have begun implementing the changes you suggested in my original code, but I have encountered a problem that I am struggling to resolve elegantly. I have found a solution mapped up in my brain but I believe there may be a way more efficient approach. Additionally, the indexing of theano tensors, s…

I see two issues mixed up in the question, one related to indexing and one related to modeling. For indexing, it seems like you mostly have everything well in hand. Actually all you are doing here is making indicator features for all the different types of dates you want to highlight. I agree the f…

hiya, could you please provide the code you used for plotting the predictions, seems like these plots changed a bit since pymc3, did you manually sort it(.mean() per chain to get it from 3d-2d and then plotting) out or do we have access to a az function that produces this directly, couldnt find any …

I didn’t use arviz. It’s the same as the code i posted here: [image] Is this half cauchy model correct? Questions The best practice for out of sample prediction is to: Use pm.MutableData objects to hold your data Include size=X.shape[0] in the likelihood term Use pm.set…

How can i sample posteriors from training-data, run posterior predictive on testing data (with complex factor graph)

Questions v5

jessegrabowski April 23, 2023, 11:05am 10

I didn’t use arviz. It’s the same as the code i posted here:

Topic		Replies	Views
Posterior predictive distributions of parameter values differs a lot from posterior? General	18	755	November 23, 2023
Get samples of posterior on new data mapped with set_data General	6	358	November 15, 2023
Best Practices for Time Series Forecasting version agnostic	11	4873	August 15, 2024
Is this half cauchy model correct? Questions modeling	28	1973	July 7, 2024
Pymc5 out of sample v5 modeling	9	1500	August 20, 2023

How can i sample posteriors from training-data, run posterior predictive on testing data (with complex factor graph)

Related topics