An initial effort has been done to include many models for the modelling and classification task. However, there are several small improvements that could greatly benefit the package.
Hyperparameters:
When doing hyper parameter tunning more information should be displayed of which hyper parameters are being tunned and what the grid search looks like.
Hyperparameter tunning should give more options on which hyper parameters to tune in two different aspects: categorical parameters should be optimised (e.g. choosing the type of kernel in SVR) and the user should be able to choose which hyperparameters to optimise and which shouldn't. Right now it optimises all of the possible hyperparameters available.
Models:
Multi Layer Perceptrons should be added as an option.
The user should be able to choose which optimizer or at least information should be displayed on which optimization strategy is being employed.
There should be an option of ensemble models where when we have several systems defined an ensemble is created that is for example the average of all the system predicted ages.
Maybe we could let the user import the model they want to use directly: by using import command (e.g. “-model from sklearn.svm import svr”
Interpretability:
As an initial approach in models in which some information can be display should. For example, in Linear Regression the weights of each feature should be printed. Something similar should be explored for each of the models available.
Classification:
Logistic regressors could also be trained when inputing several system deltas by adding or averaging all the deltas together.
We could explore Different types of classifiers and an optimum cutoff threshold for seperating should be printed out.
An initial effort has been done to include many models for the modelling and classification task. However, there are several small improvements that could greatly benefit the package.
Hyperparameters:
Models:
Interpretability:
Classification: