Model Zoo and Leaderboard#
The ModelZoo in Modeva serves as a centralized repository for managing multiple predictive models, either built-in interpretable models (to be introduced later) or wrapped external models. It is designed to simplify the process of model addition, training, and performance leaderboard.
Model Management#
from modeva import ModelZoo
from modeva.models import (
MoElasticNet, MoDecisionTreeRegressor,
MoXGBRegressor, MoRandomForestRegressor,
)
mz = ModelZoo(name="CaliforniaHousing", dataset=ds)
mz.add_model(model=MoElasticNet(name="ElasticNet", l1_ratio=0.5))
mz.add_model(model=MoDecisionTreeRegressor(name="DecisionTree"))
mz.add_model(model=MoXGBRegressor(name="XGB-Depth2", max_depth=2))
mz.add_model(model=MoRandomForestRegressor(name="RF-Depth5", max_depth=5))
Above is an example of adding four built-in models to the ModelZoo for the California Housing dataset. We can continue adding other candidate models. The ModelZoo.get_model method can be used to retrieve each individual model by its name.
Training and Leaderboard#
mz.train_all()
mz.leaderboard(order_by="test MSE", ascending=True)
Above showcases the batch training of all models in the ModelZoo, with outpupt of a performance leaderboard of all the models based on the test MSE.