CARTE: Pretraining and Transfer for Tabular Learning

This dataset is the tabular-data benchmark used in the CARTE paper (https://arxiv.org/abs/2402.16785)

CARTE is a pretrained model for tabular data by treating each table row as a star graph and training a graph transformer on top of this representation.

It has the particularity of being made of tables with high-cardinality string.

The codes for CARTE can be found at https://github.com/soda-inria/carte

Descriptions

• Anime Planet
  (https://www.kaggle.com/datasets/hernan4444/animeplanet-recommendation-database-2020)
  This dataset contains information about anime scrapped from the website Anime-Planet. The task is to predict the average rating of the anime on this site.
• Babies R Us
  (http://pages.cs.wisc.edu/~anhai/data/784_data/bikes/csv_files/babies_r_us.csv)
  Information of baby products scraped from the Babies R Us website. The task is to predict the price of baby products.
• Buy Buy Baby
  (http://pages.cs.wisc.edu/~anhai/data/784_data/bikes/csv_files/buy_buy_baby.csv)
  Information of baby products scraped from the Buy Buy Baby website. The task is to predict the price of baby products.
• Beer Ratings
  (https://www.kaggle.com/datasets/ruthgn/beer-profile-and-ratings-data-set)
  The dataset contains tasting profiles and consumer reviews for 3197 unique beers from 934 different breweries. The task is to predict overall review ratings of different beers.
• Bikedekho
  (http://pages.cs.wisc.edu/~anhai/data/784_data/bikes/csv_files/bikedekho.csv)
  Information on bikes and scooters from bikedekho website in India. The task is to predict the price of bikes.
• Bikewale
  (http://pages.cs.wisc.edu/~anhai/data/784_data/bikes/csv_files/bikewale.csv)
  Information on bikes and scooters from bikewale website in India. The task is to predict the price of bikes.
• Cardekho
  (https://www.kaggle.com/datasets/sukritchatterjee/used-cars-dataset-cardekho)
  This dataset contains information on used cars, with their listing price in the websit Cardekho. The task is to predict the price.
• Chocolate Bar Ratings
  (https://www.kaggle.com/datasets/rtatman/chocolate-bar-ratings)
  Dataset containing information and expert rating on cocoa batches. The task is to predict the rating.
• Clear Corpus
  (https://www.commonlit.org/blog/introducing-the-clear-corpus-an-open-dataset-to-advance-research-28ff8cfea84a)
  Generic information about the reading passage excerpts for elementary school students. The task is to predict the readability of the excerpts. The text feature is the name of the book, not the excerpt.
• Coffee Ratings
  (https://www.kaggle.com/datasets/hanifalirsyad/coffee-scrap-coffeereview)
  Dataset scraped from coffeereview.com containing information on various coffees. The task is to predict the review ratings of the coffees.
• Company Employees
  (https://www.kaggle.com/peopledatalabssf/free-7-million-company-dataset)
  Information on companies with over $1,000$ employees. The task is to predict the number of employees of each company.
• Employee remuneration and expenses earning over 75000
  (https://opendata.vancouver.ca/explore/dataset/employee-remuneration-and-expenses-earning-over-75000/information/?disjunctive.department&disjunctive.title)
  Remuneration and expenses for employees earning over $75,000 per year. The task is to predict the remuneration of employees.
• Employee Salaries
  (https://openml.org/d/42125)
  Information on salaries for employees of the Montgomery County, MD. The task is to predict the current annual salary range of the employees.
• Fifa22 Players
  (https://www.kaggle.com/datasets/joebeachcapital/fifa-players)
  Information on soccer players and their ability scores in Fifa22 game. The task is to predict the player's wage.
• Filmtv Movies
  (https://www.kaggle.com/datasets/stefanoleone992/filmtv-movies-dataset/data)
  Information of movies and ratings scraped from an Italian movie review website Filmtv Movies. The task is to predict the public vote on movies.
• Journal Score JCR
  (https://jcr.clarivate.com/jcr/home)
  Scientific journals and their descriptive features from Journal Citation Reports. The task is to predict the impact factors of the journals.
• Journal Score SJR
  (https://www.scimagojr.com/journalrank.php)
  Scientific journals and their descriptive features from Scimago journal rank. The task is to predict the H-index of journals.
• Japanese Anime
  (https://www.kaggle.com/datasets/dbdmobile/myanimelist-dataset)
  List of Japanese animes and their relevant information. The task is to predict score for the animes.
• K-Drama
  (https://www.kaggle.com/datasets/noorrizki/top-korean-drama-list-1500)
  List of korean drama and their basic information from mydramalist website. The task is to predict the score of the Korean dramas.
• Michelin
  (https://www.kaggle.com/datasets/ngshiheng/michelin-guide-restaurants-2021)
  List of restaurants along with additional details curated from the Michelin Restaurants guide. The task is to predict the award of the restaurants.
• ML/DS Salaries
  (https://ai-jobs.net/salaries/download/salaries.csv)
  salary and basic information of workers in machine learning and data science industry. The task is to predict the salary of workers.
• Movie Revenues
  (https://www.kaggle.com/rounakbanik/the-movies-dataset)
  Metadata of movies released on or before July 2017. The task is to predict the range of the box-office revenues.
• Museums
  (https://www.kaggle.com/datasets/markusschmitz/museums)
  General information on the US museums. The task is to predict the revenues across the museums.
• Mydramalist
  (https://www.kaggle.com/datasets/rajchinagundi/mydramalist-complete-dataset)
  General information on Asian drama scraped from mydramalist website. The task is to predict the ratings of Asian dramas.
• NBA Draft
  (https://www.kaggle.com/datasets/mattop/nba-draft-basketball-player-data-19892021)
  Information on all NBA Draft picks from 1989-2021. The task is to predict the `value over replacement' of players.
• Prescription Drugs
  (https://data.ca.gov/uk/dataset/prescription-drugs-introduced-to-market)
  The data contains new prescription drugs introduced to market in California with a Wholesale Acquisition Cost (WAC) that exceeding Medicare Part D. The task is to predict WAC at introduction.
• Ramen ratings
  (https://www.kaggle.com/datasets/ankanhore545/top-ramen-ratings-2022)
  The dataset contains ratings and characteristics of various ramens produced from multiple countries. The task is to predict the range of ratings of the ramens.
• Roger Ebert
  (https://github.com/gabrielcs/movie-ratings-prediction)
  The dataset contains movies ratings by famous critic Rogert Ebert. The task is to predict the range of ratings.
• Rotten Tomatoes
  (http://pages.cs.wisc.edu/~anhai/data/784_data/movies1/csv_files/rotten_tomatoes.csv)
  Contain information on movies that can be found in Rotten Tomatoes movie rating website. The task is to predict the rating values of the movies.
• Spotify
  (https://www.kaggle.com/datasets/maharshipandya/-spotify-tracks-dataset)
  Generic information on Spotify tracks with some associated audio features. The task is to predict the popularity of the albums.
• US Accidents
  (https://smoosavi.org/datasets/us_accidents)
  Information of accidents in US cities between 2016 and 2020. From this dataset, two tasks are conducted: (1) the range of accident counts for the US cities (2) the severity of the reported accidents.
• US Presidential
  (https://github.com/alexis-cvetkov/KEN)
  Voting statistics in the 2020 US presidential election along with information on US counties. The task is to predict the range of voting numbers across US counties.
• Used Cars 24
  (https://www.kaggle.com/datasets/avikasliwal/used-cars-price-prediction)
  Information on used cars. The task is to predict the price.
• Used Cars Benz Italy
  (https://www.kaggle.com/datasets/bogdansorin/second-hand-mercedes-benz-registered-2000-2023-ita)
  Dataset containing information on used cars sold in Italy. The task is to predict the price.
• UsedCars.com
  Dataset containing information on used cars usedcars.com. The task is to predict the price.
• Used Cars Pakistan
  (https://www.kaggle.com/datasets/mustafaimam/used-car-prices-in-pakistan-2021)
  Dataset containing information on used cars sold in Pakistan. The task is to predict the price.
• Used Cars Saudi Arabia
  (https://www.kaggle.com/datasets/turkibintalib/saudi-arabia-used-cars-dataset)
  Dataset containing information on used cars sold in Saudi Arabia from the Syarah Website. The task is to predict the price of used cars.
• Videogame Sales
  (https://www.kaggle.com/datasets/gregorut/videogamesales)
  This dataset contains a list of video games with sales greater than 100,000 copies (scrape of vgchartz.com). The task is to predict the global sales of the videogames.
• Whisky
  (https://whiskyanalysis.com/index.php/database/)
  Basic and tasting information on whiskies form the whiskyanalysis.com. The task is to predict the range of meta critic of the whiskies.
• Wikiliq
  (https://www.kaggle.com/datasets/limtis/wikiliq-dataset)
  Information on alcohol that can be found in the Wikiliq website. We conducted two tasks to predict the prices of (1) beer and (2) spirits.
• Wina Poland
  (https://www.kaggle.com/datasets/skamlo/wine-price-on-polish-market)
  Information about wines on the polish market. The task is to predict the price.
• Wine.com
  (https://www.kaggle.com/datasets/manyregression/updated-wine-enthusiast-review)
  Information on wines scraped from the wine.com website. We conducted two tasks on prediction of (1) wine ratings and (2) wine prices.
• WineEnthusiasts
  (https://www.kaggle.com/datasets/manyregression/updated-wine-enthusiast-review)
  Information about a wine and a taster from winemag.com. We conducted two tasks on prediction of (1) wine ratings and (2) wine prices.
• WineVivino
  (https://www.kaggle.com/datasets/joshuakalobbowles/vivino-wine-data)
  Information about wine bottles scrapped from Vivino's website. We conducted two tasks on prediction of (1) wine ratings and (2) wine prices.
• Yelp
  (https://www.yelp.com/dataset)
  The Yelp Open dataset for academic research. We extracted information on the restaurants from the original dataset. The task is to predict the range of stars for the restaurants.
• Zomato
  (https://www.kaggle.com/datasets/anas123siddiqui/zomato-database?select=restaurant.csv)
  Information and reviews of restaurants found in the zomato websites. The task is to predict the range of ratings for each restaurants.

Our paper

@article{kim2024carte,
 title={CARTE: pretraining and transfer for tabular learning},
 author={Kim, Myung Jun and Grinsztajn, L{\'e}o and Varoquaux, Ga{\"e}l},
 journal={arXiv preprint arXiv:2402.16785},
 year={2024}
}