Implement linear tree surrogate

[bammari]

Fixes

We generalize the OMLTSurrogate base class to include models other than Keras models. We refactor these changes across the KerasSurrogate class as well. We do this because future interfaces to OMLT models may inherity from the OMLTSurrogate base class without being specific to Keras. These are mostly semantic changes.

Summary/Motivation:

We add an interface to the OMLT linear-tree capabilities to allow users to implement Linear Model Decision Trees as surrogates within their IDAES models.

Changes proposed in this PR:

• Generalize OMLTSurrogate
• Add LinearTreeSurrogate class which inherits from OMLTSurrogate
• Add tests to test this new interface.

Legal Acknowledgement

By contributing to this software project, I agree to the following terms and conditions for my contribution:

1. I agree my contributions are submitted under the license terms described in the LICENSE.txt file at the top level of this directory.
2. I represent I am authorized to make the contributions and grant the license. If my employer has rights to intellectual property that includes these contributions, I represent that I have received permission to make contributions and grant the required license on behalf of that employer.

[bpaul4]

These changes look reasonable to me, just a couple of very minor "missing periods" in docstrings but feel free to ignore. I will update the branch and approve the workflow.

[bpaul4]

Suggested change

	with load_from_folder
	with load_from_folder.

[bpaul4]

Suggested change

	Args:
	lt_model: Linear-tree model
	This is the Linear-tree model that will be loaded.
	input_labels: list of str
	The ordered list of labels corresponding to the inputs in the linear-tree model
	output_labels: list of str
	The ordered list of labels corresponding to the outputs in the linear-tree model
	input_bounds: None of dict of tuples
	Keys correspond to each of the input labels and values are the tuples of
	bounds (lb, ub)
	input_scaler: None or OffsetScaler
	The scaler to be used for the inputs. If None, then no scaler is used
	output_scaler: None of OffsetScaler
	The scaler to be used for the outputs. If None, then no scaler is used
	Args:
	lt_model: Linear-tree model
	This is the Linear-tree model that will be loaded.
	input_labels: list of str
	The ordered list of labels corresponding to the inputs in the linear-tree model.
	output_labels: list of str
	The ordered list of labels corresponding to the outputs in the linear-tree model.
	input_bounds: None of dict of tuples
	Keys correspond to each of the input labels and values are the tuples of
	bounds (lb, ub).
	input_scaler: None or OffsetScaler
	The scaler to be used for the inputs. If None, then no scaler is used.
	output_scaler: None of OffsetScaler
	The scaler to be used for the outputs. If None, then no scaler is used.

[bpaul4]

@bammari can you please look into the pylint issue? See below:

************* Module idaes.core.surrogate.linear_tree_surrogate
idaes/core/surrogate/linear_tree_surrogate.py:35:4: E0401: Unable to import 'omlt.linear_tree' (import-error)
idaes/core/surrogate/linear_tree_surrogate.py:35:4: E0611: No name 'linear_tree' in module 'omlt' (no-name-in-module)

[bammari]

@bpaul4 Thanks for pointing this out.

It looks like Github CI installs omlt v1.1 in the idaes-pse-dev conda environment when we run actions. Linear tree capabilities were introduced in omlt v1.2.

We have specified omlt==1.1 in setup.py. The most recent version is omlt==1.2.2. I believe changing this should resolve this issue. I have included a new commit that tests this.