views-runs 1.12.3

Tools for doing model runs with views

views-runs

This package is meant to help views researchers with training models, by providing a common interface for data partitioning and stepshift model training. It also functions as a central hub package for other classes and functions used by views researchers, including stepshift (StepshiftedModels) and views_partitioning (DataPartitioner).

Installation

To install views-runs, use pip:

pip install views-runs

This also installs the vendored libraries stepshift and views_partitioning.

Usage

The library offers a class imported at views_runs.ViewsRun, that wraps the to central components of a ViEWS 3 run: A partitioning scheme expressed via a views_partitioning.DataPartitioner instance, and a stepshifted modelling process expressed via a stepshift.views.StepshiftedModels instance. For documentation on the data partitioner, see views_partitioning. For documentation on stepshifted modelling, see views.StepshiftedModels.

The wrapper takes care of applying these two classes to your data, in order to produce predictions in a familiar and predictable format, as well as ensuring that there is no overlap between training and testing partitions. Instantiating a run requires instances of both of these classes, like so:

run = ViewsRun(
   DataPartitioner({"A":{"train":(1,100),"test":(101,200)}}),
   StepshiftedModels(LogisticRegression,[1,2,3,4,5,6],"my_dependent_variable"),
)

This instance can then be applied to a time-unit indexed dataframe to train the models, and produce predictions for the timespans defined in the data partitioner:

run.fit("A","train",dataframe)
predictions = run.predict("A","test",dataframe)

Examples

There are notebooks that show various workflows with views_runs and the vendored libraries:

• BasicExample.ipynb