trainFunctions

phs.trainFunctions

Classes:

Name	Description
Experts	Convenience Class for a Mixture of Experts Classifier.
Ksplit	Convenience Class k-Split (for k-fold CV).

Functions:

Name	Description
getTrainedModel	Train a model using Linear SVM Classifier.

Experts dataclass

Convenience Class for a Mixture of Experts Classifier.

Methods:

Name	Description
__get__	Predict the class of X based on "opinion" of
__getitem__	Convenience subscript access for experts.
acc	Convenience function of naïve accuracy

Attributes:

Name	Type	Description
classes	list[int]
experts	list[dict]

classes = field(default_factory=lambda: [0, 1]) class-attribute instance-attribute

experts instance-attribute

__get__(X)

Predict the class of X based on "opinion" of experts. Y = self(X)

__getitem__(idx)

Convenience subscript access for experts.

self[i] is the same as self.experts[i]

acc(X, Y)

Convenience function of naïve accuracy calculator.

acc = np.mean(self(X)==Y)

Ksplit dataclass

Convenience Class k-Split (for k-fold CV).

Split a dataset with \(N\) samples into \(k\) parts. Only maintain indices so that ksplit[i] provides a set of indices subscriptable to the original dataset, e.g. iTrain, iVal = ksplit[i]; xTrain = X[iTrain] creates a train data subset for i-th fold of validation.

On initialisation, create a random shuffled set of indices if not already provided.

[i] subscript access will fetch the i-th pair of indices corresponding to train and val split respectively.

Attributes:

Name	Type	Description
N	int
indices	list[int]
k	int

N instance-attribute

indices = field(default_factory=list) class-attribute instance-attribute

k instance-attribute

getTrainedModel(xTrain, yTrain, xVal, yVal, hparams)

Train a model using Linear SVM Classifier.

hparams is a dict with key 'C'