{
  "fetched_at": "2026-10-19T07:38:39Z",
  "source_url": "https://www.kaggle.com/competitions/applications-of-deep-learning-wustl-fall-2024/data",
  "pages": [
    {
      "name": "data-description",
      "content": "The entire dataset consists of tabular, text, and images for 1,200 fictional homes. These homes come in variety of property types, ranging from single-family, multi-family, apartment buildings and sky-rise penthouses. Each single-family and multi-family is a unique building, but multiple condo units can belong to the same apartment or sky-rise building. Below is a detailed description of the database.\n\n## Files\n\n*   **df-train_1000x12.csv** - training data consisting of 10 default features and their home values\n*   **df-test_200x11.csv** - testing data consisting of 10 default features\n*   **SimReports/Home-*.txt** - inspection reports for all 1200 Homes\n*   **SimPictures/Home-*.png** - street-level images for all 1200 Homes\n*   **sample_submission.csv** - a sample submission file in the correct format\n\n## Columns\n\n*   `homeid` - Unique 5-digit home identifier\n*   `feat_yrbuilt` - Year the home was built\n*   `feat_nrooms` - Number of total rooms\n*   `feat_nbed` - Number of bedrooms\n*   `feat_nbath` - Number of bathrooms\n*   `feat_sqft` - Squared footage of the entire home\n*   `feat_proptype` - Whether the home is a single-family, multi-family, or condo\n*   `feat_basement` - Whether the home has a basement\n*   `feat_stories` - Whether the home has single or multiple floors \n*   `feat_solar` - Whether the home has solar panels\n*   `feat_nsales` - Number of times sold in the past decade\n*   `target_homevalue` - Home values (rounded to the nearest dollar)\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "This dataset was simulated using a combination of methods:\n\n**Tabular Data** were generated using a conditional tabular generative adversarial network and k-mean clustering from real-world data collected in 2017, which was subsequently adjusted and scaled for the purposes of this competition.\n\n**Inspection Reports** were generated using GPT-4-Turbo from OpenAI. The prompts instructed the LLM to provide a qualitative description about the internal condition of this home and a comprehensive estimate of the total repair costs.\n\n**Street-level Images** were generated using Juggernaut_final from DigiPlay, which is an open-source text-to-image model that is accessible via the Hugging Face platform API.\n\nAs is the case in real-world housing data, the home values are positively skewed, whereby a small subset of houses are much more expansive relative to the average priced home. Aside from the skewed distribution of the target variable, there are no anomalies or black swans that need to be taken in consideration. All key features are self-contained within this competition - it is not recommended to incorporate external housing data or sources of information.\n\nSimilar Kaggle Competition: https://www.kaggle.com/competitions/applications-of-gen-ai-wustl-fall-2024",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "Submissions are evaluated based on the [Mean Absolute Error (MAE)](https://www.kaggle.com/code/metric/mean-absolute-error/log?scriptVersionId=151169616), which measures the average magnitude of errors between the predicted and actual home values. A mean absolute error of 10,000 would suggests that, across all homes in the test sample, you can expect the predictions to be approximately $10,000 away from the actual home values.\n\n## Submission File\nFor each **homeid** in the test set, you must predict the **target_homevalue** variable, which reflects the dollar amount for a given house. The file should contain a header and have the following format:\n\n    homeid,target_homevalue\n    01624,123456\n    07647,654321\n    08202,99999\n    etc.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "Teams of up to 5 students are allowed. ",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "abstract",
      "content": "This competition is provided as an opportunity to implement various text and image-based methods on a relatively simple and clean dataset. Your task is to predict home values, as if you were a Data Scientist for a company like Opendoor and Offerpad.\n\nFor this challenge you will be provided descriptive data stored in tabular format, inspection reports about the internal condition of each home, and street-level images of each building. As such, this dataset contains both structured and unstructured elements that provide complementary information about home values, all of which can be used to engineer novel features and potentially improve predictive accuracy.\n\nA comprehensive approach would be to utilize the information stored in all three formats, including tabular, text, and images. There are an endless number of ways to achieve this goal. For instance, one way would be parsing through the text files to extract specific information, whereas another approach could be using word embedding models to convert all the text into a numerical representation. Beyond feature engineering, there are additional decisions that must be made about the best way to approach predictive modeling, such as how to combine these different sets of features into a single model or whether an ensemble machine learning framework would work better.\n\nThis competition allows you to explore the effectiveness of different techniques across a variety of data types, all with the common goal of deriving a machine learning solution that can accurately predict home values.\n\n[The full dataset/solution and simulation code is available through this link](https://www.kaggle.com/datasets/robertjirsaraie/predicting-home-values-from-a-land-far-far-away/data)\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F6602523%2F7ae796fc46db2d00766d1a8ff86eca83%2FHome-00490.jpg?generation=1727639534008323&alt=media)",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}