---
license: apache-2.0
task_categories:
- summarization
language:
- en
tags:
- sports
pretty_name: NBA Play-By-Play to Recap
configs:
- config_name: default
  data_files:
  - split: train
    path: "train.zip"
  - split: test
    path: "test.zip"
  - split: validation
    path: "validation.zip"
  - split: unsupervised
    path: "unsupervised.zip"
---

```
DatasetDict({
    train: Dataset({
        features: ['metadata', 'input', 'output'],
        num_rows: 4593
    })
    test: Dataset({
        features: ['metadata', 'input', 'output'],
        num_rows: 574
    })
    validation: Dataset({
        features: ['metadata', 'input', 'output'],
        num_rows: 574
    })
    unsupervised: Dataset({
        features: ['metadata', 'input', 'output'],
        num_rows: 27257
    })
})
```

This dataset consists of 3 columns:
- metadata - which has the home team, away team, and time of a single NBA game.
- input - which has a string representation of a play-by-play of the game.
- output - which has a human-written recap of the game.

The train-test-validation sets consist of all of the NBA regular season games of:
- 2019-20
- 2020-21
- 2021-22
- 2022-23
- 2023-24
with 80%, 10%, 10% ratio

The unsupervised set consists of metadata and input only (output field is an empty string) of all of the NBA regular season games of:
- 1996-97
- 1997-98
- 1998-99
- 1999-2000
- 2000-01
- 2001-02
- 2002-03
- 2003-04
- 2004-05
- 2005-06
- 2006-07
- 2007-08
- 2008-09
- 2009-10
- 2010-11
- 2011-12
- 2012-13
- 2013-14
- 2014-15
- 2015-16
- 2016-17
- 2017-18
- 2018-19