ModelTool
#179
brycefrank
started this conversation in
Ideas
`ModelTool`
#179
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
A frequent pattern in forest inventories are large sets of models that are selected by a forest analyst for routine use in predicting forest attributes. For example, a biometrician in Oregon might select models that predict merchantable volumes for all the tree species located on an ownership. Examples of software that achieve this are the NVEL published by the USFS.
The
ModelTool
class is a way to construct these sets of models. Below is some sketching about what features we wantModelTool
to have and how it connects in the other areas (e.g., databases andallometric/models
).ModelTool
s should be constructed after publications files are ingested. This allows the tool to access any model and avoids complex interdependencies.ModelTool
can contain multiple response variables (e.g., the NVEL does volumes of different types)ModelTool
is constructed usingconstructer_fn
, which selects the models out of theallometric_models.RDS
or (in the future) the MongoDB. This can potentially use model IDs, but not necessarily (because model IDs tend to be unstable).Slots
citation
- UnlikePublication
this slot is optional. This covers the case for publications that establish a model tool. One example might be the models in an FVS variant document. The variant document in this case is cited as the canonical source of the model tool.In some cases the tool is not clearly attributed to one source, hence why this is optional. For example, the Canadian biomass models (from Piotr) were sourced from two interlacing primary sources, and it appears to be common knowledge that the combined set of them are used as a model tool. In this case one citation is not appropriate and should not be defined. Instead, the user may access the component citations.
Methods
predict(tool, data)
- should return an n x p tibble containing all predictions for all n trees and all p attributessummary
- should provide a short one-paragraph blurb about the model tool, the attributes that are predicted and the data requirements for each attribute(?)Beta Was this translation helpful? Give feedback.
All reactions