{ "trained_at": "2026-04-26T08:12:01.990046+00:00", "data_run_date": "2026-04-25", "version": "v0.3-finetuned-mpnet", "description": "v0.2 + contrastive fine-tuning of mpnet on price-similarity triplets", "parent_version": "v0.2", "fine_tune": { "base_model": "sentence-transformers/all-mpnet-base-v2", "triplets": 3252, "epochs": 3, "lr": 2e-05, "batch_size": 64, "val_triplet_acc_before": 0.5185185185185185, "val_triplet_acc_after": 0.6049382716049383, "improvement_pp": 8.64197530864198 }, "splits": { "train": { "rows": 265240, "start": "2022-01-28", "end": "2023-09-30" }, "val": { "rows": 3545, "start": "2023-10-01", "end": "2023-12-31" }, "test": { "rows": 2744, "start": "2024-01-01", "end": "2024-05-04" } }, "feature_count": 146, "feature_cols": [ "len", "n_digits", "n_letters", "n_special", "n_lower", "n_upper", "is_palindrome", "is_all_digits", "is_all_letters", "is_ascii", "has_unicode", "starts_digit", "ends_digit", "max_char_run", "n_unique_chars", "in_wikipedia", "in_geonames", "in_us_firstname", "in_iso3166", "in_ticker", "in_sec_edgar", "in_wiktionary_en", "wordlist_hits", "club__social_handles", "club__prepunk_full_rankings", "club__personas", "club__gamertags", "club__top500_cities_global", "club__familynames_usa", "club__common_english", "club__top_nouns", "club__top500_cities_usa", "club__common_animals", "club__catholicism", "club__crypto_terms", "club__finance_terms", "club__pokemon_gen4", "club__logistics", "club__us_government", "club__top_crypto_names", "club__mythical_creatures", "club__firstnames_usa", "club__gen_alpha", "club__sports", "club__crayola_classic", "club__performing_arts", "club__top_crypto_tickers", "club__gamertags_double", "club__country_codes", "club__us_states", "club__pokemon_gen1", "club__conspiracy_theories", "club__wikidata_top_fantasy_char", "club__historic_figures", "club__currency_symbols", "club__luxury", "club__natural_wonders", "club__pokemon_gen3", "club__paranormal", "club__holidays", "club__memes", "club__currency_names", "club__home", "club__pokemon_gen2", "club__fine_art", "n_clubs", "trademark_conflict", "name_age_days", "prior_transfer_count", "fg_value", "eth_tvl_usd", "eth_stable_mcap", "eth_dex_volume", "nft_total_fee_usd", "pca_000", "pca_001", "pca_002", "pca_003", "pca_004", "pca_005", "pca_006", "pca_007", "pca_008", "pca_009", "pca_010", "pca_011", "pca_012", "pca_013", "pca_014", "pca_015", "pca_016", "pca_017", "pca_018", "pca_019", "pca_020", "pca_021", "pca_022", "pca_023", "pca_024", "pca_025", "pca_026", "pca_027", "pca_028", "pca_029", "pca_030", "pca_031", "pca_032", "pca_033", "pca_034", "pca_035", "pca_036", "pca_037", "pca_038", "pca_039", "pca_040", "pca_041", "pca_042", "pca_043", "pca_044", "pca_045", "pca_046", "pca_047", "pca_048", "pca_049", "pca_050", "pca_051", "pca_052", "pca_053", "pca_054", "pca_055", "pca_056", "pca_057", "pca_058", "pca_059", "pca_060", "pca_061", "pca_062", "pca_063", "knn_count", "knn_mean_log", "knn_median_log", "knn_p90_log", "knn_max_sim", "knn_min_sim", "knn_log_max", "knn_log_min" ], "pca_dim": 64, "best_iteration": 369, "xgb_params": { "objective": "reg:squarederror", "eval_metric": "rmse", "tree_method": "hist", "device": "cuda", "max_depth": 7, "learning_rate": 0.04, "subsample": 0.85, "colsample_bytree": 0.65, "min_child_weight": 8, "reg_alpha": 0.5, "reg_lambda": 2.0, "seed": 42 }, "metrics": { "train": { "r2_log": 0.7984388470649719, "rmse_log": 0.7249601483345032, "mae_log": 0.48019057512283325, "median_ape": 0.2934191823005676, "bias_log": -0.0001444444787921384 }, "val": { "r2_log": 0.6729167103767395, "rmse_log": 1.0476961135864258, "mae_log": 0.7321875691413879, "median_ape": 0.48088952898979187, "bias_log": 0.12580855190753937 }, "test": { "r2_log": 0.32547563314437866, "rmse_log": 1.527344822883606, "mae_log": 1.2592869997024536, "median_ape": 1.3042705059051514, "bias_log": 0.712732195854187 } }, "top_features": [ { "name": "knn_mean_log", "gain": 2475.822021484375 }, { "name": "knn_median_log", "gain": 1594.9427490234375 }, { "name": "knn_p90_log", "gain": 1196.794189453125 }, { "name": "in_wikipedia", "gain": 432.5325927734375 }, { "name": "len", "gain": 427.3330078125 }, { "name": "pca_002", "gain": 208.85269165039062 }, { "name": "is_all_digits", "gain": 207.7695770263672 }, { "name": "n_digits", "gain": 207.7196807861328 }, { "name": "pca_001", "gain": 198.31832885742188 }, { "name": "pca_007", "gain": 197.07334899902344 }, { "name": "n_clubs", "gain": 186.21694946289062 }, { "name": "name_age_days", "gain": 184.0680694580078 }, { "name": "knn_count", "gain": 183.48965454101562 }, { "name": "n_unique_chars", "gain": 170.0704345703125 }, { "name": "is_palindrome", "gain": 158.3007049560547 }, { "name": "eth_stable_mcap", "gain": 153.7845001220703 }, { "name": "trademark_conflict", "gain": 153.01382446289062 }, { "name": "pca_022", "gain": 145.79708862304688 }, { "name": "ends_digit", "gain": 140.9295196533203 }, { "name": "club__prepunk_full_rankings", "gain": 130.34014892578125 }, { "name": "eth_tvl_usd", "gain": 120.09010314941406 }, { "name": "n_special", "gain": 116.31814575195312 }, { "name": "n_letters", "gain": 108.61463928222656 }, { "name": "is_ascii", "gain": 108.28950500488281 }, { "name": "pca_000", "gain": 103.93241119384766 } ], "wandb_run": "https://wandb.ai/quantumly-aletheia-research/ens-appraiser/runs/0eevvvo0" }