Deserialization of RegressionModels triggers a recalcuatlion of estimated values

[pfleck]

On deserializing a gradient boosted tree solution, the GBTSurrogateModel recaltulates the actual GBTModel even if the model is not yet used - which can be quite runtime expensive.

Instead, the model should only be created when the model is actually needed to estimate values.

[pfleck]

For GBT to enable lazy loading, the Lazy<T>.IsValueCreated must be checked before checking for null in the cloning constructor. Otherwise, the check for null triggers the expensive model-creation.

if (original.actualModel.IsValueCreated && original.ActualModel != null) clonedModel = cloner.Clone(original.ActualModel);

However, this fix causes the GUI to freeze because recalculating is triggered later on the GUI-thread. In the future, a progress bar should indicate that the recalculation is ongoing.

[mkommend]

After further investigations on this issue the following conclusions can be drawn:

• Philipp is right
• Only GBT models are affected, because cloning of a GBT surrogate model trigges a recalculation of the model. The fix to check whether a value is created solves this issue, but has the described disadvantage of freezing the GUI.
• CreateEnsembleMenuItem triggers cloning of solutions whenever a view is changed to check if it should be enabled. There must be a better way with less overhead to enable /disable the menu item.
• For models that are not stored completely a special solution view should be created that allows explicitly to trigger the model recalculation and blocks access to the solution evaluation view until the recalculation is finished.