Datasets:

klokedm
/

tabnetics-runs

campaign large_stringclasses 4 values	profile large_stringclasses 278 values	dataset_shard large_stringclasses 9 values	run_timestamp large_stringlengths 0 15	dataset_id large_stringclasses 63 values	dataset_name large_stringclasses 63 values	tier large_stringclasses 4 values	effective_tier large_stringclasses 4 values	domain large_stringclasses 2 values	platform large_stringclasses 4 values	seed int64 11 97	config large_stringclasses 1 value	protocol large_stringclasses 1 value	row_type large_stringclasses 2 values	is_winner bool 2 classes	fs_method_preset large_stringclasses 43 values	compute_budget large_stringclasses 1 value	n_samples_total float64 41 7k ⌀	n_features_total float64 500 54.7k ⌀	n_train float64 32 5.6k ⌀	n_test float64 9 1.4k ⌀	model large_stringclasses 212 values	selection_strategy large_stringclasses 3 values	accuracy float64 0 1 ⌀	balanced_accuracy float64 0 1 ⌀	macro_f1 float64 0 1 ⌀	hybrid_score float64 0 1 ⌀	roc_auc float64 0 1 ⌀	roc_curve_type large_stringclasses 2 values	roc_auc_source large_stringclasses 4 values	model_cv_score float64 0.11 1 ⌀	model_cv_train_test_gap float64 -0.05 0.15 ⌀	n_features_selected float64 0 200 ⌀	n_portfolio_candidates float64 0 36 ⌀	n_clf_candidates_evaluated float64 2 24 ⌀	fs_time_sec float64 0.44 92.4k ⌀	dist_time_sec float64 0 9.39k ⌀	transform_time_sec float64 0 9.39k ⌀	dist_fit_time_sec float64 0 9.39k ⌀	fs_plus_clf_time_sec float64 4.09 92.4k ⌀	dist_transform_time_sec float64 0 9.39k ⌀	classification_stage2_wall_sec float64 0.09 4.19k ⌀	data_source large_stringclasses 63 values	batch_label_policy large_stringclasses 2 values	batch_label_policy_reason large_stringclasses 2 values	multiomics_feature_blocks_source_reason large_stringclasses 1 value	multiomics_adapter_mode large_stringclasses 2 values	multiomics_integrator large_stringclasses 1 value	multiomics_adapter_reason large_stringclasses 2 values	validation_pipeline large_stringclasses 1 value	validation_scenario float64	promotion_eligible float64 1 1 ⌀	promotion_blocker float64	source_policy large_stringclasses 2 values	max_train_samples float64	n_fs_subset float64 12 2.24k ⌀	n_dist_features_fitted float64 0 256 ⌀	n_dist_features_transformed float64 0 256 ⌀	n_dist_rejected float64 0 255 ⌀	n_dist_skipped_unreliable float64 0 60 ⌀	n_low_gof_downweighted float64 0 26 ⌀	mean_dist_stability_weight float64 1 1 ⌀	sota_holdout_bal_acc_low float64 0 0.96 ⌀	sota_holdout_bal_acc_high float64 0.49 1 ⌀	sota_inflated_bal_acc_low float64 0.5 0.99 ⌀	sota_inflated_bal_acc_high float64 0.62 1 ⌀	sota_holdout_status large_stringclasses 3 values	sota_inflated_status large_stringclasses 3 values	sota_bal_acc_low float64 0 0.96 ⌀	sota_bal_acc_high float64 0.49 1 ⌀	sota_status large_stringclasses 3 values	protocol_gap_note large_stringclasses 1 value	sanity_ok float64 1 1 ⌀	enabled_methods_source large_stringclasses 5 values	maqc_pairing_selected_fs_name float64	maqc_pairing_selected_cv_score float64	classifier_conformal_enabled float64 0 1 ⌀	classifier_conformal_applied float64 0 1 ⌀	df_stage_position large_stringclasses 2 values	df_stage_source_space large_stringclasses 2 values	classifier_conformal_skip_reason large_stringclasses 2 values	classifier_conformal_coverage float64 0.33 1 ⌀	classifier_conformal_set_size_mean float64 1 14 ⌀	classifier_conformal_singleton_rate float64 0 1 ⌀	classifier_conformal_method large_stringclasses 4 values	classifier_conformal_mapie_method large_stringclasses 4 values	classifier_conformal_mapie_skip_reason large_stringclasses 3 values	regime_policy_mode large_stringclasses 4 values	regime_policy_bypass_mode large_stringclasses 2 values	regime_policy_p_over_n float64 0.03 1.71k ⌀	fs_stability_threshold_method large_stringclasses 1 value	fs_stability_target_pfer float64 1 1 ⌀	effective_enabled_methods large_stringclasses 43 values	telemetry_effective_methods_populated float64 1 1 ⌀	batch_correction_mode_applied float64	batch_correction_apply_reason large_stringclasses 2 values	dist_criterion large_stringclasses 10 values	n_total_runs_in_shard int64 2 135	n_failures_in_shard int64 0 13	cfg_classifier_selection_mode large_stringclasses 3 values	cfg_classification_backend large_stringclasses 2 values	cfg_flaml_time_budget int64 30 120	cfg_optuna_time_budget int64 120 120	cfg_optuna_n_trials int64 25 25	cfg_classifier_oracle_k int64 1 3	cfg_classifier_oracle_weighting_mode large_stringclasses 4 values	cfg_enable_classifier_conformal bool 2 classes	cfg_classifier_conformal_alpha float64 0.1 0.1	cfg_selection_strategy large_stringclasses 2 values	cfg_fs_portfolio_size int64 6 6	cfg_enable_fs_adaptive_portfolio_sizing bool 2 classes	cfg_fs_adaptive_size_min float64 4 4 ⌀	cfg_fs_adaptive_size_max float64 8 8 ⌀	cfg_screening_enabled bool 2 classes	cfg_screening_method large_stringclasses 3 values	cfg_mnpo_performance_oracle_mode large_stringclasses 2 values	cfg_use_stability_oracle bool 2 classes	cfg_use_complexity_oracle bool 2 classes	cfg_use_robust_oracle bool 2 classes	cfg_fs_diversity_oracle_mode large_stringclasses 1 value	cfg_fs_oracle_weighting_mode large_stringclasses 4 values	cfg_df_stage_position large_stringclasses 2 values	cfg_df_family_set large_stringclasses 2 values	cfg_disable_cdf_transform bool 2 classes	cfg_disable_cdf_reliability_gate bool 1 class	cfg_enable_cdf_block_gating_cv bool 1 class	cfg_enable_df_fastpath bool 1 class	cfg_max_dist_features int64 256 256	cfg_disable_rank_prefilter bool 2 classes	cfg_prefilter_top_k int64 600 600	cfg_prefilter_strategies large_stringclasses 2 values	cfg_regime_gating_enabled bool 2 classes	cfg_regime_gating_target_tier large_stringclasses 1 value	cfg_batch_correction large_stringclasses 4 values	cfg_multiomics_adapter large_stringclasses 1 value	cfg_test_size float64 0.2 0.2	cfg_tier_lockout_enabled bool 2 classes	cfg_tier_routing_enabled bool 1 class	cfg_enable_fs_ipss_eats_threshold bool 1 class	cfg_fs_ipss_target_fdr float64 0.15 0.15	cfg_fs_copula_knockoff_draws int64 30 30	cfg_fs_copula_alpha_kn float64 0.1 0.1	config_flags_json large_stringclasses 605 values	sota_source_confidence large_stringclasses 3 values	sota_claim_scope large_stringclasses 3 values	portfolio_candidates_list large_stringclasses 8 values	classification_backend large_stringclasses 1 value	classification_selection_mode large_stringclasses 1 value	n_clf_candidates_failed float64 0 1 ⌀
val-18	C02_pool_legacy_plus_tabpfn	ds02	20260312_095938	gisette_nips03	GISETTE (NIPS 2003 FS Challenge)	medium	medium	non_genomic	cDNA	42	baseline	holdout	winner	true	mnpo_v14_core_plus_ipss	standard	7,000	5,000	5,600	1,400	tabpfn	mnpo_portfolio	0.982857	0.982857	0.982857	0.982857	0.99865	binary	predict_proba	null	null	null	null	null	20,183.145969	662.35876	662.356354	null	null	null	null	hf_bundle:klokedm/tabnetics-expanded/gisette_nips03	none	policy_none	source_modality_membership_unavailable	none	mb_plsda	disabled	fs	null	1	null	standard	null	2,240	80	17	61	2	1	1	0.75	0.89	0.85	0.97	above	within	0.75	0.89	above	Published SOTA often uses LOOCV / repeated CV / best-of-N protocols. This benchmark reports both strict-holdout and inflated-reference ranges, with strict holdout used for promotion decisions.	1	config	null	null	1	1	after_fs	prefilter_raw	null	0.981429	1	1	split	none	not_requested	none	null	0.105357	fixed	1	['gradient_boosting', 'linear_svm', 'mutual_information', 'anova_f', 'mrmr_jmi', 'ova_ensemble', 'class_pareto_front', 'boruta', 'copula_knockoff', 'decorrelated_stability', 'relieff', 'stability_lasso', 'rfecv', 'hsic_lasso', 'joint_multiclass_support', 'ipss']	1	null	mode_none	simple	2	3	legacy	sklearn	60	120	25	1	tritrust	true	0.1	mnpo_portfolio	6	true	4	8	true	evalue	multi_model_oracles	true	true	true	legacy_jaccard	banzhaf	after_fs	flex	false	false	false	false	256	false	600	["mi_ftest_blend","rf_importance","wsnr"]	true	very_hard	none	none	0.2	false	false	false	0.15	30	0.1	{"allow_synthetic_fallback":false,"dataset_integrity_policy":"error","dataset_min_classes":2,"dataset_min_class_count":1,"synthetic_sample_cap":2500,"synthetic_feature_cap":10000,"progress_heartbeat_sec":30.0,"progress_watchdog_sec":0.0,"test_size":0.2,"max_train_samples":0,"disable_df_robust":false,"disable_df_lrt":fa...	null	null	null	null	null	null
val-18	C02_pool_legacy_plus_tabpfn	ds02	20260312_095938	gisette_nips03	GISETTE (NIPS 2003 FS Challenge)	medium	medium	non_genomic	cDNA	59	baseline	holdout	winner	true	mnpo_v14_core_plus_ipss	standard	7,000	5,000	5,600	1,400	tabpfn	mnpo_portfolio	0.975	0.975	0.975	0.975	0.996706	binary	predict_proba	null	null	null	null	null	20,717.464546	609.540916	609.532503	null	null	null	null	hf_bundle:klokedm/tabnetics-expanded/gisette_nips03	none	policy_none	source_modality_membership_unavailable	none	mb_plsda	disabled	fs	null	1	null	standard	null	2,240	80	23	55	2	2	1	0.75	0.89	0.85	0.97	above	within	0.75	0.89	above	Published SOTA often uses LOOCV / repeated CV / best-of-N protocols. This benchmark reports both strict-holdout and inflated-reference ranges, with strict holdout used for promotion decisions.	1	config	null	null	1	1	after_fs	prefilter_raw	null	0.972143	1	1	split	none	not_requested	none	null	0.104821	fixed	1	['gradient_boosting', 'linear_svm', 'mutual_information', 'anova_f', 'mrmr_jmi', 'ova_ensemble', 'class_pareto_front', 'boruta', 'copula_knockoff', 'decorrelated_stability', 'relieff', 'stability_lasso', 'rfecv', 'hsic_lasso', 'joint_multiclass_support', 'ipss']	1	null	mode_none	simple	2	3	legacy	sklearn	60	120	25	1	tritrust	true	0.1	mnpo_portfolio	6	true	4	8	true	evalue	multi_model_oracles	true	true	true	legacy_jaccard	banzhaf	after_fs	flex	false	false	false	false	256	false	600	["mi_ftest_blend","rf_importance","wsnr"]	true	very_hard	none	none	0.2	false	false	false	0.15	30	0.1	{"allow_synthetic_fallback":false,"dataset_integrity_policy":"error","dataset_min_classes":2,"dataset_min_class_count":1,"synthetic_sample_cap":2500,"synthetic_feature_cap":10000,"progress_heartbeat_sec":30.0,"progress_watchdog_sec":0.0,"test_size":0.2,"max_train_samples":0,"disable_df_robust":false,"disable_df_lrt":fa...	null	null	null	null	null	null
val-18	C04_pool_mnpo_plus_tabpfn	ds02	20260313_183732	gisette_nips03	GISETTE (NIPS 2003 FS Challenge)	medium	medium	non_genomic	cDNA	11	baseline	holdout	winner	true	mnpo_v14_core_plus_ipss	standard	7,000	5,000	5,600	1,400	mnpo_lr	mnpo_portfolio	0.973571	0.973571	0.973571	0.973571	0.994945	binary	predict_proba	null	null	null	null	null	27,514.275115	592.911195	592.908663	null	null	null	null	hf_bundle:klokedm/tabnetics-expanded/gisette_nips03	none	policy_none	source_modality_membership_unavailable	none	mb_plsda	disabled	fs	null	1	null	standard	null	2,240	80	21	58	1	1	1	0.75	0.89	0.85	0.97	above	within	0.75	0.89	above	Published SOTA often uses LOOCV / repeated CV / best-of-N protocols. This benchmark reports both strict-holdout and inflated-reference ranges, with strict holdout used for promotion decisions.	1	config	null	null	1	1	after_fs	prefilter_raw	null	0.969286	1	1	split	none	not_requested	none	null	0.105357	fixed	1	['gradient_boosting', 'linear_svm', 'mutual_information', 'anova_f', 'mrmr_jmi', 'ova_ensemble', 'class_pareto_front', 'boruta', 'copula_knockoff', 'decorrelated_stability', 'relieff', 'stability_lasso', 'rfecv', 'hsic_lasso', 'joint_multiclass_support', 'ipss']	1	null	mode_none	simple	5	0	mnpo_hybrid	sklearn	60	120	25	2	tritrust	true	0.1	mnpo_portfolio	6	true	4	8	true	evalue	multi_model_oracles	true	true	true	legacy_jaccard	banzhaf	after_fs	flex	false	false	false	false	256	false	600	["mi_ftest_blend","rf_importance","wsnr"]	true	very_hard	none	none	0.2	false	false	false	0.15	30	0.1	{"allow_synthetic_fallback":false,"dataset_integrity_policy":"error","dataset_min_classes":2,"dataset_min_class_count":1,"synthetic_sample_cap":2500,"synthetic_feature_cap":10000,"progress_heartbeat_sec":30.0,"progress_watchdog_sec":0.0,"test_size":0.2,"max_train_samples":0,"disable_df_robust":false,"disable_df_lrt":fa...	null	null	null	null	null	null
val-18	C04_pool_mnpo_plus_tabpfn	ds02	20260313_183732	gisette_nips03	GISETTE (NIPS 2003 FS Challenge)	medium	medium	non_genomic	cDNA	23	baseline	holdout	winner	true	mnpo_v14_core_plus_ipss	standard	7,000	5,000	5,600	1,400	mnpo_lr	mnpo_portfolio	0.97	0.97	0.969996	0.969998	0.993071	binary	predict_proba	null	null	null	null	null	27,232.790665	575.58514	575.58301	null	null	null	null	hf_bundle:klokedm/tabnetics-expanded/gisette_nips03	none	policy_none	source_modality_membership_unavailable	none	mb_plsda	disabled	fs	null	1	null	standard	null	2,240	80	25	53	2	1	1	0.75	0.89	0.85	0.97	above	within	0.75	0.89	above	Published SOTA often uses LOOCV / repeated CV / best-of-N protocols. This benchmark reports both strict-holdout and inflated-reference ranges, with strict holdout used for promotion decisions.	1	config	null	null	1	1	after_fs	prefilter_raw	null	0.967143	1	1	split	none	not_requested	none	null	0.104821	fixed	1	['gradient_boosting', 'linear_svm', 'mutual_information', 'anova_f', 'mrmr_jmi', 'ova_ensemble', 'class_pareto_front', 'boruta', 'copula_knockoff', 'decorrelated_stability', 'relieff', 'stability_lasso', 'rfecv', 'hsic_lasso', 'joint_multiclass_support', 'ipss']	1	null	mode_none	simple	5	0	mnpo_hybrid	sklearn	60	120	25	2	tritrust	true	0.1	mnpo_portfolio	6	true	4	8	true	evalue	multi_model_oracles	true	true	true	legacy_jaccard	banzhaf	after_fs	flex	false	false	false	false	256	false	600	["mi_ftest_blend","rf_importance","wsnr"]	true	very_hard	none	none	0.2	false	false	false	0.15	30	0.1	{"allow_synthetic_fallback":false,"dataset_integrity_policy":"error","dataset_min_classes":2,"dataset_min_class_count":1,"synthetic_sample_cap":2500,"synthetic_feature_cap":10000,"progress_heartbeat_sec":30.0,"progress_watchdog_sec":0.0,"test_size":0.2,"max_train_samples":0,"disable_df_robust":false,"disable_df_lrt":fa...	null	null	null	null	null	null
val-18	C04_pool_mnpo_plus_tabpfn	ds02	20260313_183732	gisette_nips03	GISETTE (NIPS 2003 FS Challenge)	medium	medium	non_genomic	cDNA	37	baseline	holdout	winner	true	mnpo_v14_core_plus_ipss	standard	7,000	5,000	5,600	1,400	mnpo_tabpfn	mnpo_portfolio	0.982143	0.982143	0.982143	0.982143	0.998459	binary	predict_proba	null	null	null	null	null	26,045.551675	576.378419	576.375603	null	null	null	null	hf_bundle:klokedm/tabnetics-expanded/gisette_nips03	none	policy_none	source_modality_membership_unavailable	none	mb_plsda	disabled	fs	null	1	null	standard	null	2,240	80	26	52	2	0	1	0.75	0.89	0.85	0.97	above	within	0.75	0.89	above	Published SOTA often uses LOOCV / repeated CV / best-of-N protocols. This benchmark reports both strict-holdout and inflated-reference ranges, with strict holdout used for promotion decisions.	1	config	null	null	1	1	after_fs	prefilter_raw	null	0.98	1	1	split	none	not_requested	none	null	0.105179	fixed	1	['gradient_boosting', 'linear_svm', 'mutual_information', 'anova_f', 'mrmr_jmi', 'ova_ensemble', 'class_pareto_front', 'boruta', 'copula_knockoff', 'decorrelated_stability', 'relieff', 'stability_lasso', 'rfecv', 'hsic_lasso', 'joint_multiclass_support', 'ipss']	1	null	mode_none	simple	5	0	mnpo_hybrid	sklearn	60	120	25	2	tritrust	true	0.1	mnpo_portfolio	6	true	4	8	true	evalue	multi_model_oracles	true	true	true	legacy_jaccard	banzhaf	after_fs	flex	false	false	false	false	256	false	600	["mi_ftest_blend","rf_importance","wsnr"]	true	very_hard	none	none	0.2	false	false	false	0.15	30	0.1	{"allow_synthetic_fallback":false,"dataset_integrity_policy":"error","dataset_min_classes":2,"dataset_min_class_count":1,"synthetic_sample_cap":2500,"synthetic_feature_cap":10000,"progress_heartbeat_sec":30.0,"progress_watchdog_sec":0.0,"test_size":0.2,"max_train_samples":0,"disable_df_robust":false,"disable_df_lrt":fa...	null	null	null	null	null	null
val-18	C04_pool_mnpo_plus_tabpfn	ds02	20260313_183732	gisette_nips03	GISETTE (NIPS 2003 FS Challenge)	medium	medium	non_genomic	cDNA	42	baseline	holdout	winner	true	mnpo_v14_core_plus_ipss	standard	7,000	5,000	5,600	1,400	mnpo_lr	mnpo_portfolio	0.977143	0.977143	0.977143	0.977143	0.99601	binary	predict_proba	null	null	null	null	null	26,270.927336	578.240439	578.234749	null	null	null	null	hf_bundle:klokedm/tabnetics-expanded/gisette_nips03	none	policy_none	source_modality_membership_unavailable	none	mb_plsda	disabled	fs	null	1	null	standard	null	2,240	80	16	61	3	0	1	0.75	0.89	0.85	0.97	above	within	0.75	0.89	above	Published SOTA often uses LOOCV / repeated CV / best-of-N protocols. This benchmark reports both strict-holdout and inflated-reference ranges, with strict holdout used for promotion decisions.	1	config	null	null	1	1	after_fs	prefilter_raw	null	0.975	1	1	split	none	not_requested	none	null	0.105357	fixed	1	['gradient_boosting', 'linear_svm', 'mutual_information', 'anova_f', 'mrmr_jmi', 'ova_ensemble', 'class_pareto_front', 'boruta', 'copula_knockoff', 'decorrelated_stability', 'relieff', 'stability_lasso', 'rfecv', 'hsic_lasso', 'joint_multiclass_support', 'ipss']	1	null	mode_none	simple	5	0	mnpo_hybrid	sklearn	60	120	25	2	tritrust	true	0.1	mnpo_portfolio	6	true	4	8	true	evalue	multi_model_oracles	true	true	true	legacy_jaccard	banzhaf	after_fs	flex	false	false	false	false	256	false	600	["mi_ftest_blend","rf_importance","wsnr"]	true	very_hard	none	none	0.2	false	false	false	0.15	30	0.1	{"allow_synthetic_fallback":false,"dataset_integrity_policy":"error","dataset_min_classes":2,"dataset_min_class_count":1,"synthetic_sample_cap":2500,"synthetic_feature_cap":10000,"progress_heartbeat_sec":30.0,"progress_watchdog_sec":0.0,"test_size":0.2,"max_train_samples":0,"disable_df_robust":false,"disable_df_lrt":fa...	null	null	null	null	null	null
val-18	C04_pool_mnpo_plus_tabpfn	ds02	20260313_183732	gisette_nips03	GISETTE (NIPS 2003 FS Challenge)	medium	medium	non_genomic	cDNA	59	baseline	holdout	winner	true	mnpo_v14_core_plus_ipss	standard	7,000	5,000	5,600	1,400	mnpo_lr	mnpo_portfolio	0.960714	0.960714	0.960714	0.960714	0.991612	binary	predict_proba	null	null	null	null	null	25,297.137508	565.297016	565.285822	null	null	null	null	hf_bundle:klokedm/tabnetics-expanded/gisette_nips03	none	policy_none	source_modality_membership_unavailable	none	mb_plsda	disabled	fs	null	1	null	standard	null	2,240	80	24	54	2	3	1	0.75	0.89	0.85	0.97	above	within	0.75	0.89	above	Published SOTA often uses LOOCV / repeated CV / best-of-N protocols. This benchmark reports both strict-holdout and inflated-reference ranges, with strict holdout used for promotion decisions.	1	config	null	null	1	1	after_fs	prefilter_raw	null	0.959286	1	1	split	none	not_requested	none	null	0.104821	fixed	1	['gradient_boosting', 'linear_svm', 'mutual_information', 'anova_f', 'mrmr_jmi', 'ova_ensemble', 'class_pareto_front', 'boruta', 'copula_knockoff', 'decorrelated_stability', 'relieff', 'stability_lasso', 'rfecv', 'hsic_lasso', 'joint_multiclass_support', 'ipss']	1	null	mode_none	simple	5	0	mnpo_hybrid	sklearn	60	120	25	2	tritrust	true	0.1	mnpo_portfolio	6	true	4	8	true	evalue	multi_model_oracles	true	true	true	legacy_jaccard	banzhaf	after_fs	flex	false	false	false	false	256	false	600	["mi_ftest_blend","rf_importance","wsnr"]	true	very_hard	none	none	0.2	false	false	false	0.15	30	0.1	{"allow_synthetic_fallback":false,"dataset_integrity_policy":"error","dataset_min_classes":2,"dataset_min_class_count":1,"synthetic_sample_cap":2500,"synthetic_feature_cap":10000,"progress_heartbeat_sec":30.0,"progress_watchdog_sec":0.0,"test_size":0.2,"max_train_samples":0,"disable_df_robust":false,"disable_df_lrt":fa...	null	null	null	null	null	null
val-18	C04_pool_mnpo_plus_tabpfn	ds01	20260310_223759	xena_tcga_brca	TCGA-BRCA Breast Cancer (Xena, PAM50 5-subtype)	hard	hard	genomics	cDNA	11	baseline	holdout	winner	true	mnpo_v14_core_plus_ipss	standard	956	20,530	764	192	mnpo_elastic_net_lr	mnpo_portfolio	0.895833	0.893148	0.89485	0.893829	0.988481	ovr_micro	predict_proba	null	null	null	null	null	8,921.739407	16.514496	16.513977	null	null	null	null	hf_bundle:klokedm/tabnetics-expanded/xena_tcga_brca	none	policy_none	source_modality_membership_unavailable	none	mb_plsda	disabled	fs	null	1	null	standard	null	305	100	97	3	0	0	1	0.55	0.72	0.65	0.85	above	above	0.55	0.72	above	Published SOTA often uses LOOCV / repeated CV / best-of-N protocols. This benchmark reports both strict-holdout and inflated-reference ranges, with strict holdout used for promotion decisions.	1	config	null	null	1	1	after_fs	prefilter_raw	null	0.90625	1.03125	0.96875	split	none	not_requested	none	null	0.78534	fixed	1	['gradient_boosting', 'linear_svm', 'mutual_information', 'anova_f', 'mrmr_jmi', 'ova_ensemble', 'class_pareto_front', 'boruta', 'copula_knockoff', 'decorrelated_stability', 'relieff', 'stability_lasso', 'rfecv', 'hsic_lasso', 'joint_multiclass_support', 'ipss']	1	null	mode_none	simple	5	0	mnpo_hybrid	sklearn	60	120	25	2	tritrust	true	0.1	mnpo_portfolio	6	true	4	8	true	evalue	multi_model_oracles	true	true	true	legacy_jaccard	banzhaf	after_fs	flex	false	false	false	false	256	false	600	["mi_ftest_blend","rf_importance","wsnr"]	true	very_hard	none	none	0.2	false	false	false	0.15	30	0.1	{"allow_synthetic_fallback":false,"dataset_integrity_policy":"error","dataset_min_classes":2,"dataset_min_class_count":1,"synthetic_sample_cap":2500,"synthetic_feature_cap":10000,"progress_heartbeat_sec":30.0,"progress_watchdog_sec":0.0,"test_size":0.2,"max_train_samples":0,"disable_df_robust":false,"disable_df_lrt":fa...	null	null	null	null	null	null
val-18	C04_pool_mnpo_plus_tabpfn	ds01	20260310_223759	xena_tcga_brca	TCGA-BRCA Breast Cancer (Xena, PAM50 5-subtype)	hard	hard	genomics	cDNA	23	baseline	holdout	winner	true	mnpo_v14_core_plus_ipss	standard	956	20,530	764	192	mnpo_lr	mnpo_portfolio	0.880208	0.890849	0.86647	0.881097	0.983648	ovr_micro	predict_proba	null	null	null	null	null	8,596.191588	46.565367	46.564607	null	null	null	null	hf_bundle:klokedm/tabnetics-expanded/xena_tcga_brca	none	policy_none	source_modality_membership_unavailable	none	mb_plsda	disabled	fs	null	1	null	standard	null	305	100	97	2	1	0	1	0.55	0.72	0.65	0.85	above	above	0.55	0.72	above	"Published SOTA often uses LOOCV / repeated CV / best-of-N protocols. This benchmark reports both st(...TRUNCATED)	1	config	null	null	1	1	after_fs	prefilter_raw	null	0.901042	1.067708	0.932292	split	none	not_requested	none	null	0.78534	fixed	1	"['gradient_boosting', 'linear_svm', 'mutual_information', 'anova_f', 'mrmr_jmi', 'ova_ensemble', 'c(...TRUNCATED)	1	null	mode_none	simple	5	0	mnpo_hybrid	sklearn	60	120	25	2	tritrust	true	0.1	mnpo_portfolio	6	true	4	8	true	evalue	multi_model_oracles	true	true	true	legacy_jaccard	banzhaf	after_fs	flex	false	false	false	false	256	false	600	["mi_ftest_blend","rf_importance","wsnr"]	true	very_hard	none	none	0.2	false	false	false	0.15	30	0.1	"{\"allow_synthetic_fallback\":false,\"dataset_integrity_policy\":\"error\",\"dataset_min_classes\":(...TRUNCATED)	null	null	null	null	null	null
val-18	C04_pool_mnpo_plus_tabpfn	ds01	20260310_223759	xena_tcga_brca	TCGA-BRCA Breast Cancer (Xena, PAM50 5-subtype)	hard	hard	genomics	cDNA	37	baseline	holdout	winner	true	mnpo_v14_core_plus_ipss	standard	956	20,530	764	192	mnpo_lr	mnpo_portfolio	0.869792	0.878448	0.872086	0.875904	0.987336	ovr_micro	predict_proba	null	null	null	null	null	6,976.215712	4.738265	4.737583	null	null	null	null	hf_bundle:klokedm/tabnetics-expanded/xena_tcga_brca	none	policy_none	source_modality_membership_unavailable	none	mb_plsda	disabled	fs	null	1	null	standard	null	305	100	100	0	0	0	1	0.55	0.72	0.65	0.85	above	above	0.55	0.72	above	"Published SOTA often uses LOOCV / repeated CV / best-of-N protocols. This benchmark reports both st(...TRUNCATED)	1	config	null	null	1	1	after_fs	prefilter_raw	null	0.885417	1.005208	0.994792	split	none	not_requested	none	null	0.78534	fixed	1	"['gradient_boosting', 'linear_svm', 'mutual_information', 'anova_f', 'mrmr_jmi', 'ova_ensemble', 'c(...TRUNCATED)	1	null	mode_none	simple	5	0	mnpo_hybrid	sklearn	60	120	25	2	tritrust	true	0.1	mnpo_portfolio	6	true	4	8	true	evalue	multi_model_oracles	true	true	true	legacy_jaccard	banzhaf	after_fs	flex	false	false	false	false	256	false	600	["mi_ftest_blend","rf_importance","wsnr"]	true	very_hard	none	none	0.2	false	false	false	0.15	30	0.1	"{\"allow_synthetic_fallback\":false,\"dataset_integrity_policy\":\"error\",\"dataset_min_classes\":(...TRUNCATED)	null	null	null	null	null	null

End of preview. Expand in Data Studio

Tabnetics Validation Runs

Per-run, per-dataset, per-seed, per-method experimental results from the tabnetics automated tabular-classification pipeline across validation campaigns val-18 through val-21.

Statistic	Value
Rows	140,403
Columns	176
Campaigns	val-18, val-19, val-20, val-21
Unique datasets	63
Unique pipeline profiles	278
Seeds	11, 23, 37, 42, 59, 67, 73, 89, 97
Winner rows	56,702
Classifier-candidate rows	83,701

Row types

Every row represents one (dataset, seed, classifier) trial:

winner — the pipeline's final selected classifier for that (dataset, seed, profile) run. These rows carry full holdout metrics (accuracy, balanced_accuracy, macro_f1, hybrid_score, roc_auc, etc.) plus timing and feature-selection details from the CSV results.
classifier_candidate — a non-winning classifier from the model cross-validation stage. These rows carry the CV score (model_cv_score) and train-test gap (model_cv_train_test_gap) but not holdout metrics, since only the winner was evaluated on the held-out test set.

Key columns

Identity & metadata

Column	Description
`campaign`	Validation campaign (val-18 through val-21)
`profile`	Pipeline profile name
`dataset_shard`	Dataset shard identifier (ds0, ds1, etc.)
`run_timestamp`	Run timestamp
`dataset_id`	OpenML / internal dataset identifier
`dataset_name`	Human-readable dataset name
`tier` / `effective_tier`	Dataset complexity tier
`domain`	Dataset domain
`seed`	Random seed
`row_type`	`winner` or `classifier_candidate`
`is_winner`	Boolean flag

Performance metrics (winner rows)

Column	Description
`accuracy`	Holdout accuracy
`balanced_accuracy`	Holdout balanced accuracy
`macro_f1`	Holdout macro-F1
`hybrid_score`	Composite hybrid score
`roc_auc`	ROC-AUC (various curve types)

Model selection (all rows)

Column	Description
`model`	Classifier name
`model_cv_score`	Cross-validation score during model selection
`model_cv_train_test_gap`	CV train-test gap (overfitting indicator)

Feature selection

Column	Description
`selection_strategy`	FS strategy used
`n_features_selected`	Number of features after selection
`n_portfolio_candidates`	Size of the FS portfolio
`fs_method_preset`	FS method preset name

Timing

Column	Description
`fs_time_sec`	Feature-selection wall time
`dist_time_sec`	Distribution fitting wall time
`classification_stage2_wall_sec`	Classification stage-2 wall time

Configuration flags (`cfg_*` columns)

~40 boolean/string configuration flags from each run's metadata, prefixed with cfg_. These capture the exact pipeline settings for reproducibility. The raw JSON is also available in config_flags_json.

Usage

from datasets import load_dataset

ds = load_dataset("klokedm/tabnetics-runs", split="train")
df = ds.to_pandas()

# Winners only
winners = df[df["row_type"] == "winner"]

# All classifier candidates for a specific dataset
cands = df[(df["dataset_name"] == "wdbc") & (df["seed"] == 42)]

# Compare campaigns
df.groupby("campaign")["balanced_accuracy"].mean()