TabArena Benchmark Results

Context: TabArena is a NeurIPS 2025 benchmark suite of 38 classification datasets within a larger living tabular benchmark. This page positions the current tabnetics general run against the current TabArena leaderboard using the same core scoring model: task-weighted pairwise battles, MLE Elo with bootstrap confidence intervals, average rank, win-rate, MRR, and normalized score.

Note: this is an informational snapshot showing how tabnetics competes on general tabular data. General-tabular optimization is not currently the focus of the library, and this page should be read as a reference point rather than as a formal leaderboard submission.

Interactive browser: The same public snapshot is also available in the static Results Browser alongside the HDLSS benchmark explorer.

General profile

Run configuration

Parameter Value
Profile general
Seeds 42
Max training samples 50000
Task timeout 3600 s
Workers 12
Classifier oracle MNPO hybrid
Leaderboard bootstrap 200 rounds

Current run snapshot

On the 38 classification datasets currently covered by the merged general profile run, tabnetics (general) receives Elo 1012.1 in the overall leaderboard-style comparison, with normalized score 0.105.

The corresponding binary and multiclass leaderboard rows are shown below, followed by the dataset-level results from this run against the current official benchmark table.

Tabnetics row

method Elo Elo 95% CI Score Rank Winrate MRR
tabnetics (general) 1012.1 +105/-121 0.105 32.56 0.283 0.116

Nearby overall leaderboard rows

method Elo Score Rank Winrate
NN_TORCH (default) 1079.2 0.017 29.68 0.348
FASTAI (default) 1040.5 0.038 31.39 0.309
tabnetics (general) 1012.1 0.105 32.56 0.283
RF (default) 1000 0.022 33.04 0.272
LR (tuned + ensemble) 980.6 0.027 33.78 0.255

Binary leaderboard row

method Elo Elo 95% CI Score Rank Winrate MRR
tabnetics (general) 1008.3 +135/-151 0.076 32.84 0.276 0.099

Multiclass leaderboard row

method Elo Elo 95% CI Score Rank Winrate MRR
tabnetics (general) 1027.3 +269/-481 0.213 31.49 0.307 0.176

Per-dataset comparison against the current official best method

Dataset problem_type Bal. Acc. Tabnetics metric_error Best official metric_error Best official method Delta vs best Dataset rank Selected model
APSFailure binary 0.957 0.0139 0.0071 TABICL (default) 0.0069 42 mnpo_lr
Amazon_employee_access binary 0.756 0.1437 0.1168 CAT (tuned) 0.0269 11 mnpo_rf
Bank_Customer_Churn binary 0.766 0.1589 0.1256 TABPFNV2 (tuned) 0.0333 39 mnpo_lr
Bioresponse binary 0.788 0.1459 0.1243 XGB (tuned + ensemble) 0.0216 36 mnpo_rf
Diabetes130US binary 0.573 0.39 0.3277 GBM (tuned + ensemble) 0.0623 41 mnpo_lr
E-CommereShippingData binary 0.688 0.2715 0.2557 TABPFNV2 (default) 0.0159 42 mnpo_lr
Fitness_Club binary 0.735 0.1911 0.1781 TABPFNV2 (default) 0.013 32 mnpo_lr
GiveMeSomeCredit binary 0.71 0.2281 0.1329 TABM (tuned + ensemble) 0.0953 42 mnpo_lr
HR_Analytics_Job_Change_of_Data_Scientists binary 0.73 0.2208 0.1947 TABICL (default) 0.0261 42 mnpo_rf
Is-this-a-good-customer binary 0.678 0.298 0.2495 EBM (default) 0.0485 41 mnpo_lr
Marketing_Campaign binary 0.784 0.1342 0.0806 TABPFNV2 (tuned + ensemble) 0.0536 42 mnpo_lr
NATICUSdroid binary 0.931 0.0199 0.0126 TABICL (default) 0.0074 40 mnpo_lr
bank-marketing binary 0.697 0.2395 0.2344 CAT (default) 0.0051 27 mnpo_lr
blood-transfusion-service-center binary 0.734 0.2189 0.2445 FASTAI (tuned + ensemble) -0.0256 1 mnpo_lr
churn binary 0.856 0.0934 0.0695 MNCA (default) 0.0238 38 mnpo_xgb
coil2000_insurance_policies binary 0.632 0.3008 0.2268 TABPFNV2 (tuned + ensemble) 0.0739 40 mnpo_lr
credit-g binary 0.661 0.2452 0.2037 GBM (tuned + ensemble) 0.0416 42 mnpo_lr
credit_card_clients_default binary 0.695 0.271 0.2121 TABICL (default) 0.0589 42 mnpo_nb
customer_satisfaction_in_airline binary 0.939 0.0138 0.0049 REALMLP (tuned + ensemble) 0.0089 36 mnpo_rf
diabetes binary 0.805 0.1137 0.1556 TABPFNV2 (default) -0.0419 1 mnpo_lr
hazelnut-spread-contaminant-detection binary 0.927 0.0244 0.0076 TABDPT (default) 0.0168 22 mnpo_lgbm
heloc binary 0.715 0.2051 0.1987 TABPFNV2 (tuned + ensemble) 0.0064 27 mnpo_lr
in_vehicle_coupon_recommendation binary 0.75 0.1748 0.1483 TABM (tuned + ensemble) 0.0265 23 mnpo_lgbm
jm1 binary 0.654 0.2904 0.2239 TABICL (default) 0.0665 44 mnpo_lr
kddcup09_appetency binary 0.74 0.1867 0.1542 CAT (default) 0.0325 25 mnpo_lr
online_shoppers_intention binary 0.821 0.0996 0.0627 TABPFNV2 (tuned + ensemble) 0.0369 42 mnpo_lr
polish_companies_bankruptcy binary 0.777 0.0681 0.0187 TABPFNV2 (tuned + ensemble) 0.0494 31 mnpo_lgbm
qsar-biodeg binary 0.865 0.0822 0.0615 TABICL (default) 0.0207 40 mnpo_lr
seismic-bumps binary 0.651 0.2824 0.2166 TABICL (default) 0.0658 42 mnpo_lr
taiwanese_bankruptcy_prediction binary 0.837 0.0993 0.0547 REALMLP (tuned + ensemble) 0.0446 42 mnpo_lr
MIC multiclass 0.373 2.2882 0.4303 TABM (tuned + ensemble) 1.858 45 mnpo_elastic_net_lr
SDSS17 multiclass 0.964 0.1221 0.0723 RF (tuned + ensemble) 0.0498 37 mnpo_rf
anneal multiclass 0.894 0.1004 0.0156 TABPFNV2 (default) 0.0847 42 mnpo_lr
hiva_agnostic multiclass 0.349 1.43 0.1738 RF (tuned) 1.2562 45 mnpo_knn
maternal_health_risk multiclass 0.84 0.3628 0.4048 TABDPT (default) -0.0419 1 mnpo_xgb
splice multiclass 0.975 0.1048 0.0993 TABPFNV2 (tuned + ensemble) 0.0056 5 mnpo_xgb
students_dropout_and_academic_success multiclass 0.693 0.6439 0.5266 TABPFNV2 (tuned + ensemble) 0.1173 42 mnpo_lr
website_phishing multiclass 0.885 0.2802 0.2215 TABPFNV2 (tuned + ensemble) 0.0587 32 mnpo_rf

Interpretation

This page is intended as an informational view of the current tabnetics run against the current TabArena leaderboard. It provides a general-tabular reference point while the library remains focused primarily on HDLSS problems.

Documentation and webpages on this site are generated from authoritative internal sources using a combination of deterministic rules and generative AI. Errors are possible. Please report issues via GitHub Discussions or email [email protected].