Metadata-Version: 2.1
Name: views-runs
Version: 1.13.0
Summary: Tools for doing model runs with views
Home-page: https://www.github.com/prio-data/views_runs
Author: peder2911
Author-email: pglandsverk@gmail.com
Requires-Python: >=3.8,<3.10
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.8
Classifier: Programming Language :: Python :: 3.9
Requires-Dist: scikit-learn (>=1.0.2,<2.0.0)
Requires-Dist: stepshift (>=2.2.1,<3.0.0)
Requires-Dist: views-partitioning (>=3.0.0,<4.0.0)
Requires-Dist: views-schema (>=2.2.0,<3.0.0)
Requires-Dist: viewser (>=5.12.0,<6.0.0)
Description-Content-Type: text/markdown

# 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](https://github.com/prio-data/stepshift) (StepshiftedModels)
and [views_partitioning](https://github.com/prio-data/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](https://www.github.com/prio-data/views_partitioning). For documentation on stepshifted modelling, see 
[views.StepshiftedModels](https://github.com/prio-data/viewser/wiki/Stepshift).


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](https://github.com/prio-data/viewser/wiki/DataConventions#time-unit-indexed-pandas-dataframes)
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](examples/BasicExample.ipynb)

