-
-
Notifications
You must be signed in to change notification settings - Fork 2k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Imputation does not work in combination with pm.Data #4441
Comments
Is this something that might be problematic if automatically inputed (setting new Data with more/less missing values than the initial data)? |
The whole implementation for switching the data in an existing model is broken. I see |
I have a hunch that we'll have to revisit the whole imputation feature under the new Let's label this "wontfix" and revisit it for PyMC3 |
Agreed 👌 |
This way I can use the |
After learning about imputed variables, this feature would require a considerable change in the internals, since all the imputation logic is happening during the model definition. |
Yes, we can't have support for the combination of SharedVariable+imputation. |
Isn't the default observed ConstantData? |
Observed are not automatically tracked with dims/coords and don't show up in model_to_graphviz. |
I see... should we check for nans in |
Let's implement Could give us some speed-up too, because with constant data the shape is known.. |
By passing `pm.Data(mutable=False)` one can create a `TensorConstant` instead of a `SharedVariable`. Data variables with known, fixed shape can enhance performance and compatibility in some situations. `pm.ConstantData` or `pm.MutableData` wrappers are provided as alternative syntax. This is the basis for solving pymc-devs#4441.
By passing `pm.Data(mutable=False)` one can create a `TensorConstant` instead of a `SharedVariable`. Data variables with known, fixed shape can enhance performance and compatibility in some situations. `pm.ConstantData` or `pm.MutableData` wrappers are provided as alternative syntax. This is the basis for solving #4441.
I think this can be closed since one can now use |
I think imputation will still fail with |
Oh because we're not yet checking |
Description
Even after #4439 imputations don't work in combination with
pm.Data
.This is because
pm.Data
creates aSharedVariable
that currently does not support anp.ma.MaskedArray
.Almost identical to the example from #4437:
Please provide the full traceback.
Versions and main components
The text was updated successfully, but these errors were encountered: