A code generation dataset for generating the code that implements Hearthstone and Magic The Gathering card effects.
This dataset contains the language to code datasets described in our paper:
There are two datasets in the paper.
Both contain card descriptions and the code that implements them. These are obtained from open source implementations of each game (acknowledgements to the authors and contributors of these toolkits).
The hearthstone dataset was crawled from: https://github.com/danielyule/hearthbreaker
The magic dataset was crawled from: https://github.com/magefree/mage
Both datasets are placed in the /third_party folder with their respective license files. The below is a description of all files:
The data (./hearthstone/carddatahs.txt,./magic/carddatamagic.txt) is organized in multiple lines as follows: for HS: -Name -Cost -Type -Rarity -Race -Class -Description -Health -Attack -Durability -Number of lines in code -Code
for Magic -Name -Cost -Type -Rarity -Id (needed as it is used in the code) -Booster it comes from -Description -Power -Toughness -Loyalty (there are these avatar cards that have these) -Number of liens in code -Code
These can be split using the splits used in the paper in the split file (./hearthstone/splitshs.txt, ./magic/splitsmagic.txt), where each line indicates which set each sample was placed (0->train, 1->dev, 2->test).
We also provide the train, dev and test splits on a per line basis.
The .in files contain a per line card descriptions separated by Keywords, such as "NAME_END", which describes that the previous tokens are the name of the card.
The .out files contain the code of the each line in the respective .in file, (\n are replaced by the § token).
For any questions regarding this dataset email:
Wang Ling [email protected]