Model version

Model version design

For each model, you can create different model versions that are based on specific choices of algorithm and features.

A model version is always associated to a model. The choice of features available for designing and training a model version is conditioned by the set of features of the TD set associated to the model. The results of the research done at feature selection level can also be used to reduce the number of features required to train your model.

Data for model version training

The data set required to train your model version is automatically extracted at run time in your training notebook. The features to be extracted will be automatically gathered from your model version metadata.

Model version training notebook

Model training is performed through the instantiation of training notebooks that are written using the TranzAI SDK.

You can directly instantiate pre-built notebooks provided by Tranzai's data science team. If you want to design your own notebooks, you have to design them with the TranzAI SDK and declare your custom notebooks in your project environment.