Hello everyone. I’ve been checking various related posts but so far still unable to perform sample_posterior_predictive. I have defined a model constructor function def model_ordered_logistic_partial_pooled(coords, cohort=cohort_): with pm.Model(coords=coords) as model: x_0 = pm…

Ignore that you have two models, and focus just on the predictive model. If you are creating some variables and indexing, you have to make sure those are consistent with each other. If you create a variable with ten entries and have an index with a value of 20 you are going to get an index error. Y…

Slowly losing hope; the additional model for prediction is still showing TypeError: ("The type's shape ((7447,)) is not compatible with the data's ((5949,))", 'Container name "b0"') where 7447 is the overall number of groups in the full data set; 5949 is the number of groups in the training model. …

I suggest you try to start with the simplest case and expand gradually. That should help you pinpoint the problem faster. Here is a simplified example: import pymc as pm idx = [0, 0, 0, 1, 1, 1] coords = { "id": [0, 1], "obs_idx": list(range(len(idx))), } with pm.Model(coords=coords) as m…

[image] ricardoV94: import pymc as pm idx = [0, 0, 0, 1, 1, 1] coords = { "id": [0, 1], "obs_idx": list(range(len(idx))), } with pm.Model(coords=coords) as m: x = pm.Data("x", [0, 0, 0, 0, 0, 0], dims=("obs_idx",)) mu_group = pm.Normal("mu_group") b = pm.Normal("b", mu_group…

There was a small typo in my example (edited). Here it is running on Google Colab: Google Colab If you are using an old version of PyMC do update it if you can.

Thanks. I noticed that the new idx has to be [0, 0, 1], instead of say [9, 9, 10]. Is there a particular reason for this requirement?

Update: I think I’ve seen someone else making the same observation, that it turns out sample_posterior_predictive works only if the number of out-of-sample forecast is the same as training set.

That’s not the case, but if you are using dims, you may need to set predictions=True so InferenceData doesn’t complain. Do you have an example?

num_cohort = 2 data_list = [-1.2083030673141326, 0.5652466185040567, -0.7851719239971027, 0.20513500717041422, 0.5652466185040567, -0.1549766041632283, 2.0056930638386268, -1.2983309701475432, 0.745302424170878, 1.1054140355045206, -0.42506031266346017, -0.6951440211636921, -1.280325389…

temp_df2 = testing_data_lite idx2 = temp_df2.race_idx Doesn’t look like you are using idx2 in the prediction model but still idx. Certainly you are doing some indexing error that has nothing to do with sample_posterior_predictive

How to properly do out-of-sample prediction for hierarchical model

Questions v5

ricardoV94 August 29, 2023, 12:31pm 10

You would need the array to have 9 dummy entries if you wand idx values of 9 and 10 to work

Out of sample/model predictions in hierarchical model with new observed data

Topic		Replies	Views
Out of sample/model predictions in hierarchical model with new observed data v5 modeling	9	525	October 16, 2024
Out of Sample Predictions on heirarchical regression model (PyMC - v5.0.1) v5	5	630	March 2, 2023
How do we predict on new unseen groups in a hierarchical model in PyMC3? Questions	10	5503	September 12, 2022
Out of sample prediction with new category version agnostic	5	996	August 3, 2022
Pymc5 out of sample v5 modeling	9	1363	August 20, 2023

How to properly do out-of-sample prediction for hierarchical model

Related topics