GeometriX Model Studio

Build models whose behavior can be understood, not merely measured

An integrated environment for interpretable model development: data analysis, a model zoo designed for transparency, calibration, and explanation of models built anywhere.

Get Model Studio

The workflow

1 · Data

Load, profile and prepare data: exploratory analysis, data-quality summaries, feature engineering and selection.

2 · Models

Train inherently interpretable models or wrap external ones — with hyperparameter tuning and pipelines.

3 · Understand

Inherent interpretation for transparent models, post-hoc explanation for the rest, and calibration throughout.

A model zoo built for transparency

Interpretable families alongside wrapped industry-standard learners, so challenger comparisons live in one place.

Interpretable by design	Wrapped & comparable
GAMI-Net · GLM Tree · GLM Tree Boost · Neural Tree · Decision Tree · Linear/GLM · ReLU-DNN · DirectRS · ICL-MoE · GBDT Leaf Kernel · FuseKernel · Mixture of Experts	XGBoost · LightGBM · CatBoost · Random Forest · Gradient Boosting · your own models via wrappers

From data to explanation in a dozen lines

Here is what the code below actually does, in plain terms. It loads a public dataset of hourly bike rentals, prints a profile of the data, and draws the distribution of the number we want to predict. Then it trains two models on the same data: a standard gradient-boosting model of the kind most teams use today, and a mixture-of-experts model built to stay interpretable. Two lines each. From there, the same objects feed directly into explanation and testing.

from modeva import DataSet
from modeva.models import MoLGBMRegressor, MoMoERegressor

ds = DataSet()
ds.load("BikeSharing")
ds.summary()
ds.eda_1d(feature="cnt", plot_type="density")

model = MoLGBMRegressor(name="LGBM", max_depth=2, n_estimators=100)
model.fit(ds.train_x, ds.train_y.ravel())

moe = MoMoERegressor(name="MOE", max_depth=2, n_clusters=5, n_estimators=100)
moe.fit(ds.train_x, ds.train_y.ravel())

From the published BikeSharing example notebook

Then hand it to Assurance

Everything built (or wrapped) in Model Studio is immediately testable in Model Assurance — same DataSet, same wrappers, same result objects.

Model Assurance →

Get Model Studio today

Model Studio ships inside the modeva package. Install it and everything below works now.

$ pip install modeva

from modeva import DataSet
from modeva.models import MoLGBMRegressor, MoMoERegressor