Skip to main content
CARLA HQ
Real Estate & Urban Planning Finance & Banking use casesNACE L68RegressionTarget: median_house_value_usdOpenML #537 ↗

Houses

The California Houses dataset contains 20,640 census block group records from the 1990 U.S. Census, structured to predict median house values using demographic, spatial, and structural real estate features.

✨ Try in CarlaInstall Chrome Extension ↗100% In-Browser WebGPU • Zero Cloud Upload
20,640Records (Rows)
8Predictive Features
9 / 0Numeric / Categorical
0.0%Missing Value Ratio
40,971RMSE
50.25sWebGPU Mean Split
BUSINESS CONTEXT & OBJECTIVEReal Estate & Property Valuation (L68)

Business Objective: Houses

01Business Context

Automated valuation models (AVMs) and property investment firms require granular, location-sensitive appraisals to assess neighborhood residential asset values accurately without manual site inspections.

02Analytical Objective

Predict the median residential house value for geographic block groups based on spatial coordinates, household demographics, and structural housing metrics.

03Economic & Decision Impact

Accurate property valuation mitigates underwriting risk in residential mortgage lending, eliminates appraisal latency, and prevents severe capital misallocation stemming from under- or over-valued real estate portfolios.

ML Benchmark Narrative

Tree-based ensembles such as LightGBM, XGBoost, and CatBoost historically establish strong baselines on the California Houses dataset by capturing non-linear geographical interactions between latitude and longitude. Zero-shot tabular foundation models like TabICL and Carla match or exceed fine-tuned gradient-boosted trees in few-shot in-context settings without requiring hyperparameter tuning or feature engineering.

Source Origin:StatLib / 1990 U.S. Census
Creator:R. Kelley Pace and Ronald Barry (1997)
License:Public Domain
FEATURE SPECIFICATION & DATA DICTIONARY

Target & Feature Column Definitions

Exhaustive business definitions, measurement units, roles, and target variables across all 9 columns.

ColVariable / Feature NameRoleTypeUnit / ScaleDescription & Business Meaning
Amedian_house_value_usdMedian House ValueTARGETNUMUSD
Median house value for households within the census block group
Bmedian_income_tens_of_thousandsMedian IncomeFeatureNUM10k USD
Median household income within the census block group (expressed in tens of thousands of USD)
Chousing_median_age_yearsMedian Housing AgeFeatureNUMyears
Median age of residential buildings within the census block group
Dtotal_rooms_countTotal RoomsFeatureNUMcount
Total count of rooms across all residential units in the block group
Etotal_bedrooms_countTotal BedroomsFeatureNUMcount
Total count of bedrooms across all residential units in the block group
Fpopulation_countPopulationFeatureNUMpeople
Total number of permanent residents residing within the census block group
Ghouseholds_countHouseholdsFeatureNUMhouseholds
Total number of households (occupying housing units) in the block group
Hlatitude_degLatitudeFeatureNUMdegrees
Centroid latitude coordinate of the census block group
Ilongitude_degLongitudeFeatureNUMdegrees
Centroid longitude coordinate of the census block group
DATASET PREVIEW

Interactive Data Table

Explore rows, feature values, and target labels for Houses.

Loading dataset...
LOCAL EXECUTION TELEMETRY

TabICLv2 WebGPU Benchmark Results

Complete performance metrics on persisted IID or non-IID evaluation splits, executed 100% locally in the browser sandbox.

Mean Split Latency?Mean wall-clock execution time per persisted evaluation split running 100% locally via WebGPU inside Chrome.Explore Latency Guide →
50.25s9 splits • 8 ensembles/split
Total Evaluation Runtime?Total wall-clock duration across all selected splits, including in-context encoding, WebGPU shader execution, and result aggregation.Explore Split Protocol →
452.27sIncludes warmup & sync
Evaluation Protocol?Precomputed row indices ensure the browser and Python runners evaluate identical out-of-sample observations without leakage.Explore Split Protocol →
3×3 SplitsRepeated IID • 9 total
TabICLv2 Score?Primary task performance score (RMSE) achieved by TabICLv2 across out-of-sample evaluation splits.Explore Metric Formula →
40,971RMSE
Evaluation SplitTrain RowsTest Rows
Split Duration?Wall-clock inference time taken for this evaluation split running 100% locally via WebGPU.Learn more →
R² Score?Coefficient of determination measuring variance explained by the model.Learn more →
MAE?Mean Absolute Error in original target units, robust to outliers.Learn more →
RMSE?Root Mean Squared Error, penalizing large residual deviations.Learn more →
MAPE?Mean Absolute Percentage Error relative to true target values.Learn more →
RMSLE?Root Mean Squared Logarithmic Error, measuring proportional error scales.Learn more →
R1 / F113,7606,88050.33s0.872725,805.5941,603.12
R1 / F213,7606,88049.97s0.860426,207.9442,515.54
R1 / F313,7606,88050.03s0.863326,266.7642,789.29
R2 / F113,7606,88049.99s0.870425,515.9341,399.63
R2 / F213,7606,88050.31s0.864626,425.7242,777.59
R2 / F313,7606,88050.45s0.869425,463.2641,516.92
R3 / F113,7606,88050.46s0.872025,459.1640,838.46
R3 / F213,7606,88050.41s0.857326,925.3943,289.24
R3 / F313,7606,88050.31s0.864725,859.7943,191.18
Mean ± Std50.25s0.8661 ± 0.005025,992.242,213.4
UNIFIED BENCHMARK LEADERBOARD

Foundation Models vs Traditional ML Baselines

Side-by-side evaluation of our in-browser Carla engine (WebGPU), open-source tabular foundation models, and OpenML baselines.

RankAlgorithm / ModelModel FamilyRuntime
R² Score?Coefficient of determination (R²) measuring proportion of variance explained.Google ML Guide & Details →
MAE?Mean Absolute Error measuring average prediction error magnitude.Learn more →
RMSE?Root Mean Squared Error penalizing large prediction errors.Learn more →
Reference / Repo
#1Google TabFM v1.0Tabular Foundation ModelPyTorch/Python0.894422,351.137,481google-research/tabfm ↗
#2TabPFN v3Tabular Foundation ModelPyTorch/Python0.882623,749.739,522.7PriorLabs/tabpfn ↗
#3EXAONE TabularTabular Foundation ModelPyTorch/Python0.880024,044.739,970.9LGAI-Research/EXAONE-Tabular ↗
#4Full TabICLv2 (PyTorch Reference)Tabular Foundation ModelPyTorch/Python0.873924,947.340,970.9soda-inria/tabicl ↗
#5nanotabicl VanillaTabular Foundation ModelPyTorch/Python0.873724,965.641,005.3soda-inria/nanotabicl ↗
#6Streaming nanotabiclTabular Foundation ModelPyTorch/Python0.871725,242.741,314.9soda-inria/nanotabicl ↗
#7Carla Engine (TabICLv2 WebGPU)Tabular Foundation ModelBrowser0.866125,992.242,213.4100% In-Browser
STATISTICAL DISTRIBUTIONS

Variable Schema & Summary Distributions

Observed numerical ranges, category cardinalities, missing rates, and sample values across 20,640 rows.

ColVariable NameTypeMissingDistinctSummary Stats / DistributionSample Values
Amedian_house_value_usdTARGETNUM03,842[14999, 500001] μ=206855.8 σ=115392.8452600358500352100
Bmedian_income_tens_of_thousandsNUM012,928[0.4999, 15.0001] μ=3.9 σ=1.98.32528.30147.2574
Chousing_median_age_yearsNUM052[1, 52] μ=28.6 σ=12.6412152
Dtotal_rooms_countNUM05,926[2, 39320] μ=2635.8 σ=2181.688070991467
Etotal_bedrooms_countNUM01,928[1, 6445] μ=537.9 σ=421.21291106190
Fpopulation_countNUM03,888[3, 35682] μ=1425.5 σ=1132.43222401496
Ghouseholds_countNUM01,815[1, 6082] μ=499.5 σ=382.31261138177
Hlatitude_degNUM0862[32.54, 41.95] μ=35.6 σ=2.137.8837.8637.85
Ilongitude_degNUM0844[-124.35, -114.31] μ=-119.6 σ=2.0-122.23-122.22-122.24
FREQUENTLY ASKED QUESTIONS

Frequently Asked Questions: Houses

Common questions regarding the Houses dataset, machine learning task formulations, and in-browser tabular inference.

What is the Houses dataset used for?

The Houses dataset is a classic tabular regression benchmark used to predict median residential home values in California based on 1990 census spatial and demographic indicators.

Can I run zero-server predictions on this dataset in Google Sheets?

Yes, Carla HQ enables zero-server in-context tabular prediction directly in spreadsheets using TabICL foundation models.

What machine learning models perform best on Houses?

Gradient-boosted decision trees (XGBoost, LightGBM, CatBoost) and tabular foundation models like TabICL perform best, capturing complex non-linear spatial relationships across latitude and longitude coordinates.

LIVE EVALUATION IN GOOGLE SHEETS

Test TabICLv2 on Houses Yourself

Open the pre-loaded Google Sheet and let Carla configure the target and task for local, zero-cloud tabular machine learning.

Step 1

Launch Carla

Click once to open the spreadsheet and Carla side panel together.

Step 2

Review the Setup

Carla selects the dataset target and task from this page automatically.

Step 3

Evaluate & Predict

Run predictions and compute metrics with zero server uploads.