model_builder
Model hyperparameters.
ModelDecisionGraphParameters
#
Hyperparameters that influence the decision graph of a Xpdeep model.
The model decision graph is characterized by a set of learnt decision and can be adjusted to the use cases to provide more interpretable explanations.
Please look at the documentation Model -> Create a model for a detailed explanation of each parameter.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
max_decision_seq_length |
In [1, 10], the maximum length of a decision sequence in the graph. If None, the parameter will be estimated depending on the task and nature of the data. |
required | |
target_homogeneity_pruning_threshold |
In [0,1]. If the target homogeneity in a decision population is above this threshold, the decision is pruned. If None, the threshold will be estimated depending on the task and nature of the data. |
required | |
population_pruning_threshold |
In [0,1], decision that have a population smaller than this threshold (as a percentage of the training set) are pruned. If None, the threshold will be estimated depending on the task and nature of the data. |
required | |
prune_step |
Pruning will be performed (if possible) every "prune_step" epochs. If None, will prune at total_epochs/2 epochs. |
required | |
target_homogeneity_weight |
Coefficient that incentivizes the model to learn a decision graph with homogeneous decision. Combined with pruning, helps to build a more concise decision graph, can hinder performance if too high. If None, the coefficient will be estimated depending on the task and nature of the data. |
required | |
discrimination_weight |
float | None
|
Coefficient that pushes the model to learn more discriminative decisions, can hinder performance if too high. If None, the coefficient will be estimated depending on the task and nature of the data. |
None
|
balancing_weight |
Coefficient that pushes the model to learn decisions with balanced population, can hinder performance if too high. |
required |