Training Functions

sklearn-classifer

Run any scikit-learn compatible classifier or list of classifiers

steps

1. generate a scikit-learn model configuration using the model_pkg_class parameter
   • input a package and class name, for example, sklearn.linear_model.LogisticRegression
   • mlrun will find the class and instantiate a copy using default parameters
   • You can modify both the model class instantiator and the fit methods (other functions could be similarly modified)
2. get a sample of data from a data source
   • select all rows using -1
   • select a random sample of rows using a negative integer
   • select consecutive rows using a positive integer
3. split the data into train, validation, and test sets
   • the test set is saved as an artifact and never seen again until testing
   • WIP: this will be parametrized to produce cross-validator splits (one way of performing CV)
4. train the model
5. pickle / serialize the model
   • models can be pickled or saved as json
6. evaluate the model
   • a custom evaluator can be provided, see function doc for details