v0.7 wash filtering + recalibration: 2026-04-27
Browse files- v0_7_metadata.json +526 -0
v0_7_metadata.json
ADDED
|
@@ -0,0 +1,526 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"trained_at": "2026-04-27T08:32:28.319255+00:00",
|
| 3 |
+
"data_run_date": "2026-04-25",
|
| 4 |
+
"llm_run_date": "2026-04-26",
|
| 5 |
+
"version": "v0.7-wash-recalibration",
|
| 6 |
+
"description": "v0.6 features + wash-trade detection/down-weighting + dropped tail correction + conformal quantile recalibration",
|
| 7 |
+
"parent_version": "v0.6",
|
| 8 |
+
"changes_from_parent": [
|
| 9 |
+
"Wash-trade detection (round-trip \u22647d, sub-minute resale, wallet concentration >70%)",
|
| 10 |
+
"Down-weight wash rows in training (weight=0.1)",
|
| 11 |
+
"DROPPED tail-correction stacking (overfit on v0.6, test \u0394 R\u00b2 = -0.027)",
|
| 12 |
+
"Conformal quantile recalibration via val set"
|
| 13 |
+
],
|
| 14 |
+
"embedders": [
|
| 15 |
+
"mpnet-finetuned (from v0.3)",
|
| 16 |
+
"BAAI/bge-base-en-v1.5"
|
| 17 |
+
],
|
| 18 |
+
"splits": {
|
| 19 |
+
"train": {
|
| 20 |
+
"rows": 265240,
|
| 21 |
+
"start": "2022-01-28",
|
| 22 |
+
"end": "2023-09-30"
|
| 23 |
+
},
|
| 24 |
+
"val": {
|
| 25 |
+
"rows": 3545,
|
| 26 |
+
"start": "2023-10-01",
|
| 27 |
+
"end": "2023-12-31"
|
| 28 |
+
},
|
| 29 |
+
"test": {
|
| 30 |
+
"rows": 2744,
|
| 31 |
+
"start": "2024-01-01",
|
| 32 |
+
"end": "2024-05-04"
|
| 33 |
+
}
|
| 34 |
+
},
|
| 35 |
+
"feature_count": 212,
|
| 36 |
+
"feature_cols": [
|
| 37 |
+
"pca_000",
|
| 38 |
+
"pca_001",
|
| 39 |
+
"pca_002",
|
| 40 |
+
"pca_003",
|
| 41 |
+
"pca_004",
|
| 42 |
+
"pca_005",
|
| 43 |
+
"pca_006",
|
| 44 |
+
"pca_007",
|
| 45 |
+
"pca_008",
|
| 46 |
+
"pca_009",
|
| 47 |
+
"pca_010",
|
| 48 |
+
"pca_011",
|
| 49 |
+
"pca_012",
|
| 50 |
+
"pca_013",
|
| 51 |
+
"pca_014",
|
| 52 |
+
"pca_015",
|
| 53 |
+
"pca_016",
|
| 54 |
+
"pca_017",
|
| 55 |
+
"pca_018",
|
| 56 |
+
"pca_019",
|
| 57 |
+
"pca_020",
|
| 58 |
+
"pca_021",
|
| 59 |
+
"pca_022",
|
| 60 |
+
"pca_023",
|
| 61 |
+
"pca_024",
|
| 62 |
+
"pca_025",
|
| 63 |
+
"pca_026",
|
| 64 |
+
"pca_027",
|
| 65 |
+
"pca_028",
|
| 66 |
+
"pca_029",
|
| 67 |
+
"pca_030",
|
| 68 |
+
"pca_031",
|
| 69 |
+
"pca_032",
|
| 70 |
+
"pca_033",
|
| 71 |
+
"pca_034",
|
| 72 |
+
"pca_035",
|
| 73 |
+
"pca_036",
|
| 74 |
+
"pca_037",
|
| 75 |
+
"pca_038",
|
| 76 |
+
"pca_039",
|
| 77 |
+
"pca_040",
|
| 78 |
+
"pca_041",
|
| 79 |
+
"pca_042",
|
| 80 |
+
"pca_043",
|
| 81 |
+
"pca_044",
|
| 82 |
+
"pca_045",
|
| 83 |
+
"pca_046",
|
| 84 |
+
"pca_047",
|
| 85 |
+
"pca_048",
|
| 86 |
+
"pca_049",
|
| 87 |
+
"pca_050",
|
| 88 |
+
"pca_051",
|
| 89 |
+
"pca_052",
|
| 90 |
+
"pca_053",
|
| 91 |
+
"pca_054",
|
| 92 |
+
"pca_055",
|
| 93 |
+
"pca_056",
|
| 94 |
+
"pca_057",
|
| 95 |
+
"pca_058",
|
| 96 |
+
"pca_059",
|
| 97 |
+
"pca_060",
|
| 98 |
+
"pca_061",
|
| 99 |
+
"pca_062",
|
| 100 |
+
"pca_063",
|
| 101 |
+
"len",
|
| 102 |
+
"n_digits",
|
| 103 |
+
"n_letters",
|
| 104 |
+
"n_special",
|
| 105 |
+
"n_lower",
|
| 106 |
+
"n_upper",
|
| 107 |
+
"is_palindrome",
|
| 108 |
+
"is_all_digits",
|
| 109 |
+
"is_all_letters",
|
| 110 |
+
"is_ascii",
|
| 111 |
+
"has_unicode",
|
| 112 |
+
"starts_digit",
|
| 113 |
+
"ends_digit",
|
| 114 |
+
"max_char_run",
|
| 115 |
+
"n_unique_chars",
|
| 116 |
+
"in_wikipedia",
|
| 117 |
+
"in_geonames",
|
| 118 |
+
"in_us_firstname",
|
| 119 |
+
"in_iso3166",
|
| 120 |
+
"in_ticker",
|
| 121 |
+
"in_sec_edgar",
|
| 122 |
+
"in_wiktionary_en",
|
| 123 |
+
"wordlist_hits",
|
| 124 |
+
"club__prepunk_full_rankings",
|
| 125 |
+
"club__historic_figures",
|
| 126 |
+
"club__gamertags",
|
| 127 |
+
"club__firstnames_usa",
|
| 128 |
+
"club__top500_cities_global",
|
| 129 |
+
"club__familynames_usa",
|
| 130 |
+
"club__social_handles",
|
| 131 |
+
"club__country_codes",
|
| 132 |
+
"club__sports",
|
| 133 |
+
"club__crypto_terms",
|
| 134 |
+
"club__common_english",
|
| 135 |
+
"club__top500_cities_usa",
|
| 136 |
+
"club__top_nouns",
|
| 137 |
+
"club__home",
|
| 138 |
+
"club__conspiracy_theories",
|
| 139 |
+
"club__us_government",
|
| 140 |
+
"club__performing_arts",
|
| 141 |
+
"club__gamertags_double",
|
| 142 |
+
"club__mythical_creatures",
|
| 143 |
+
"club__finance_terms",
|
| 144 |
+
"club__catholicism",
|
| 145 |
+
"club__natural_wonders",
|
| 146 |
+
"club__fine_art",
|
| 147 |
+
"club__personas",
|
| 148 |
+
"club__luxury",
|
| 149 |
+
"club__pokemon_gen1",
|
| 150 |
+
"club__memes",
|
| 151 |
+
"club__top_crypto_tickers",
|
| 152 |
+
"club__currency_symbols",
|
| 153 |
+
"club__common_animals",
|
| 154 |
+
"club__logistics",
|
| 155 |
+
"club__currency_names",
|
| 156 |
+
"club__wikidata_top_fantasy_char",
|
| 157 |
+
"club__top_crypto_names",
|
| 158 |
+
"club__gen_alpha",
|
| 159 |
+
"club__pokemon_gen3",
|
| 160 |
+
"club__holidays",
|
| 161 |
+
"club__pokemon_gen2",
|
| 162 |
+
"club__paranormal",
|
| 163 |
+
"club__crayola_classic",
|
| 164 |
+
"club__pokemon_gen4",
|
| 165 |
+
"club__us_states",
|
| 166 |
+
"n_clubs",
|
| 167 |
+
"trademark_conflict",
|
| 168 |
+
"name_age_days",
|
| 169 |
+
"prior_transfer_count",
|
| 170 |
+
"fg_value",
|
| 171 |
+
"eth_tvl_usd",
|
| 172 |
+
"eth_stable_mcap",
|
| 173 |
+
"eth_dex_volume",
|
| 174 |
+
"nft_total_fee_usd",
|
| 175 |
+
"fame_score",
|
| 176 |
+
"crypto_relevance_ord",
|
| 177 |
+
"brand_collision_risk_ord",
|
| 178 |
+
"kind__concept",
|
| 179 |
+
"kind__random",
|
| 180 |
+
"kind__brand",
|
| 181 |
+
"kind__surname",
|
| 182 |
+
"kind__first_name",
|
| 183 |
+
"kind__abbreviation",
|
| 184 |
+
"kind__place",
|
| 185 |
+
"kind__other",
|
| 186 |
+
"kind__unknown",
|
| 187 |
+
"origin__english",
|
| 188 |
+
"origin__none",
|
| 189 |
+
"origin__mixed",
|
| 190 |
+
"origin__spanish",
|
| 191 |
+
"origin__german",
|
| 192 |
+
"origin__french",
|
| 193 |
+
"origin__japanese",
|
| 194 |
+
"origin__chinese",
|
| 195 |
+
"origin__italian",
|
| 196 |
+
"origin__slavic",
|
| 197 |
+
"origin__korean",
|
| 198 |
+
"origin__arabic",
|
| 199 |
+
"origin__other",
|
| 200 |
+
"origin__unknown",
|
| 201 |
+
"desc_pca_000",
|
| 202 |
+
"desc_pca_001",
|
| 203 |
+
"desc_pca_002",
|
| 204 |
+
"desc_pca_003",
|
| 205 |
+
"desc_pca_004",
|
| 206 |
+
"desc_pca_005",
|
| 207 |
+
"desc_pca_006",
|
| 208 |
+
"desc_pca_007",
|
| 209 |
+
"desc_pca_008",
|
| 210 |
+
"desc_pca_009",
|
| 211 |
+
"desc_pca_010",
|
| 212 |
+
"desc_pca_011",
|
| 213 |
+
"desc_pca_012",
|
| 214 |
+
"desc_pca_013",
|
| 215 |
+
"desc_pca_014",
|
| 216 |
+
"desc_pca_015",
|
| 217 |
+
"desc_pca_016",
|
| 218 |
+
"desc_pca_017",
|
| 219 |
+
"desc_pca_018",
|
| 220 |
+
"desc_pca_019",
|
| 221 |
+
"desc_pca_020",
|
| 222 |
+
"desc_pca_021",
|
| 223 |
+
"desc_pca_022",
|
| 224 |
+
"desc_pca_023",
|
| 225 |
+
"desc_pca_024",
|
| 226 |
+
"desc_pca_025",
|
| 227 |
+
"desc_pca_026",
|
| 228 |
+
"desc_pca_027",
|
| 229 |
+
"desc_pca_028",
|
| 230 |
+
"desc_pca_029",
|
| 231 |
+
"desc_pca_030",
|
| 232 |
+
"desc_pca_031",
|
| 233 |
+
"knnmp_count",
|
| 234 |
+
"knnmp_mean_log",
|
| 235 |
+
"knnmp_median_log",
|
| 236 |
+
"knnmp_p90_log",
|
| 237 |
+
"knnmp_max_sim",
|
| 238 |
+
"knnmp_min_sim",
|
| 239 |
+
"knnmp_log_max",
|
| 240 |
+
"knnmp_log_min",
|
| 241 |
+
"knnbg_count",
|
| 242 |
+
"knnbg_mean_log",
|
| 243 |
+
"knnbg_median_log",
|
| 244 |
+
"knnbg_p90_log",
|
| 245 |
+
"knnbg_max_sim",
|
| 246 |
+
"knnbg_min_sim",
|
| 247 |
+
"knnbg_log_max",
|
| 248 |
+
"knnbg_log_min"
|
| 249 |
+
],
|
| 250 |
+
"pca_dim_concat": 64,
|
| 251 |
+
"pca_dim_description": 32,
|
| 252 |
+
"name_kind_values": [
|
| 253 |
+
"concept",
|
| 254 |
+
"random",
|
| 255 |
+
"brand",
|
| 256 |
+
"surname",
|
| 257 |
+
"first_name",
|
| 258 |
+
"abbreviation",
|
| 259 |
+
"place",
|
| 260 |
+
"other",
|
| 261 |
+
"unknown"
|
| 262 |
+
],
|
| 263 |
+
"cultural_origin_values": [
|
| 264 |
+
"english",
|
| 265 |
+
"none",
|
| 266 |
+
"mixed",
|
| 267 |
+
"spanish",
|
| 268 |
+
"german",
|
| 269 |
+
"french",
|
| 270 |
+
"japanese",
|
| 271 |
+
"chinese",
|
| 272 |
+
"italian",
|
| 273 |
+
"slavic",
|
| 274 |
+
"korean",
|
| 275 |
+
"arabic",
|
| 276 |
+
"other",
|
| 277 |
+
"unknown"
|
| 278 |
+
],
|
| 279 |
+
"wash_filtering": {
|
| 280 |
+
"n_round_trip_labels": 23049,
|
| 281 |
+
"n_sub_minute_labels": 1222,
|
| 282 |
+
"n_high_conc_labels": 26255,
|
| 283 |
+
"n_total_flagged_labels": 39995,
|
| 284 |
+
"wash_train_weight": 0.1,
|
| 285 |
+
"train_wash_pct": 41.393832001206455
|
| 286 |
+
},
|
| 287 |
+
"best_xgb_params": {
|
| 288 |
+
"tree_method": "hist",
|
| 289 |
+
"device": "cuda",
|
| 290 |
+
"seed": 42,
|
| 291 |
+
"max_depth": 11,
|
| 292 |
+
"learning_rate": 0.00607909965562535,
|
| 293 |
+
"subsample": 0.5242971482550959,
|
| 294 |
+
"colsample_bytree": 0.5178101703841028,
|
| 295 |
+
"colsample_bylevel": 0.5676418253037602,
|
| 296 |
+
"min_child_weight": 3,
|
| 297 |
+
"reg_alpha": 4.318424044857591,
|
| 298 |
+
"reg_lambda": 1.7796015928106095,
|
| 299 |
+
"gamma": 2.982764564516846
|
| 300 |
+
},
|
| 301 |
+
"optuna": {
|
| 302 |
+
"n_trials": 30,
|
| 303 |
+
"best_val_rmse": 1.016706943511963,
|
| 304 |
+
"best_params": {
|
| 305 |
+
"max_depth": 11,
|
| 306 |
+
"learning_rate": 0.00607909965562535,
|
| 307 |
+
"subsample": 0.5242971482550959,
|
| 308 |
+
"colsample_bytree": 0.5178101703841028,
|
| 309 |
+
"colsample_bylevel": 0.5676418253037602,
|
| 310 |
+
"min_child_weight": 3,
|
| 311 |
+
"reg_alpha": 4.318424044857591,
|
| 312 |
+
"reg_lambda": 1.7796015928106095,
|
| 313 |
+
"gamma": 2.982764564516846
|
| 314 |
+
},
|
| 315 |
+
"best_trial": 28,
|
| 316 |
+
"warm_started_with_v0_6_best": true
|
| 317 |
+
},
|
| 318 |
+
"quantile_models": {
|
| 319 |
+
"q05": {
|
| 320 |
+
"best_iteration": 1527,
|
| 321 |
+
"best_val_rmse": 1.9560544421029673
|
| 322 |
+
},
|
| 323 |
+
"q50": {
|
| 324 |
+
"best_iteration": 1834,
|
| 325 |
+
"best_val_rmse": 1.045858812580417
|
| 326 |
+
},
|
| 327 |
+
"q95": {
|
| 328 |
+
"best_iteration": 3355,
|
| 329 |
+
"best_val_rmse": 2.1802895661452273
|
| 330 |
+
}
|
| 331 |
+
},
|
| 332 |
+
"calibration": {
|
| 333 |
+
"method": "additive",
|
| 334 |
+
"target_coverage": 0.9,
|
| 335 |
+
"delta_lower_additive": 0.08095530420541763,
|
| 336 |
+
"delta_upper_additive": 0.0,
|
| 337 |
+
"multiplicative_factor": 1.239099800588292,
|
| 338 |
+
"val_coverage_after": 0.8595204513399154
|
| 339 |
+
},
|
| 340 |
+
"metrics": {
|
| 341 |
+
"final": {
|
| 342 |
+
"train": {
|
| 343 |
+
"r2_log": 0.7497193813323975,
|
| 344 |
+
"rmse_log": 0.8078381419181824,
|
| 345 |
+
"mae_log": 0.47278380393981934,
|
| 346 |
+
"median_ape": 0.23963597416877747,
|
| 347 |
+
"bias_log": -0.1186392605304718
|
| 348 |
+
},
|
| 349 |
+
"val": {
|
| 350 |
+
"r2_log": 0.6740628480911255,
|
| 351 |
+
"rmse_log": 1.0458588600158691,
|
| 352 |
+
"mae_log": 0.7022704482078552,
|
| 353 |
+
"median_ape": 0.413220077753067,
|
| 354 |
+
"bias_log": -0.09871374070644379
|
| 355 |
+
},
|
| 356 |
+
"test": {
|
| 357 |
+
"r2_log": 0.4626653790473938,
|
| 358 |
+
"rmse_log": 1.363203525543213,
|
| 359 |
+
"mae_log": 1.0832252502441406,
|
| 360 |
+
"median_ape": 0.9336060285568237,
|
| 361 |
+
"bias_log": 0.4047386348247528
|
| 362 |
+
}
|
| 363 |
+
},
|
| 364 |
+
"coverage": {
|
| 365 |
+
"train": {
|
| 366 |
+
"coverage_90pct": 0.920309908007842,
|
| 367 |
+
"median_interval_log": 2.7887799739837646,
|
| 368 |
+
"median_interval_ratio": 16.261168645984977
|
| 369 |
+
},
|
| 370 |
+
"val": {
|
| 371 |
+
"coverage_90pct": 0.8595204513399154,
|
| 372 |
+
"median_interval_log": 3.658205986022949,
|
| 373 |
+
"median_interval_ratio": 38.79168757902037
|
| 374 |
+
},
|
| 375 |
+
"test": {
|
| 376 |
+
"coverage_90pct": 0.8083090379008746,
|
| 377 |
+
"median_interval_log": 4.073790550231934,
|
| 378 |
+
"median_interval_ratio": 58.77934691322367
|
| 379 |
+
}
|
| 380 |
+
}
|
| 381 |
+
},
|
| 382 |
+
"top_features": [
|
| 383 |
+
{
|
| 384 |
+
"name": "len",
|
| 385 |
+
"gain": 58.87883758544922
|
| 386 |
+
},
|
| 387 |
+
{
|
| 388 |
+
"name": "knnmp_mean_log",
|
| 389 |
+
"gain": 46.72057342529297
|
| 390 |
+
},
|
| 391 |
+
{
|
| 392 |
+
"name": "knnmp_median_log",
|
| 393 |
+
"gain": 41.08598327636719
|
| 394 |
+
},
|
| 395 |
+
{
|
| 396 |
+
"name": "is_all_digits",
|
| 397 |
+
"gain": 38.29785919189453
|
| 398 |
+
},
|
| 399 |
+
{
|
| 400 |
+
"name": "knnmp_count",
|
| 401 |
+
"gain": 34.5153923034668
|
| 402 |
+
},
|
| 403 |
+
{
|
| 404 |
+
"name": "knnmp_p90_log",
|
| 405 |
+
"gain": 31.491409301757812
|
| 406 |
+
},
|
| 407 |
+
{
|
| 408 |
+
"name": "in_wikipedia",
|
| 409 |
+
"gain": 30.60625457763672
|
| 410 |
+
},
|
| 411 |
+
{
|
| 412 |
+
"name": "knnmp_log_max",
|
| 413 |
+
"gain": 24.936614990234375
|
| 414 |
+
},
|
| 415 |
+
{
|
| 416 |
+
"name": "is_ascii",
|
| 417 |
+
"gain": 24.386781692504883
|
| 418 |
+
},
|
| 419 |
+
{
|
| 420 |
+
"name": "ends_digit",
|
| 421 |
+
"gain": 24.157560348510742
|
| 422 |
+
},
|
| 423 |
+
{
|
| 424 |
+
"name": "n_digits",
|
| 425 |
+
"gain": 24.11026382446289
|
| 426 |
+
},
|
| 427 |
+
{
|
| 428 |
+
"name": "has_unicode",
|
| 429 |
+
"gain": 23.687318801879883
|
| 430 |
+
},
|
| 431 |
+
{
|
| 432 |
+
"name": "name_age_days",
|
| 433 |
+
"gain": 22.8796443939209
|
| 434 |
+
},
|
| 435 |
+
{
|
| 436 |
+
"name": "origin__none",
|
| 437 |
+
"gain": 22.28803062438965
|
| 438 |
+
},
|
| 439 |
+
{
|
| 440 |
+
"name": "eth_stable_mcap",
|
| 441 |
+
"gain": 22.02800941467285
|
| 442 |
+
},
|
| 443 |
+
{
|
| 444 |
+
"name": "pca_002",
|
| 445 |
+
"gain": 21.55613136291504
|
| 446 |
+
},
|
| 447 |
+
{
|
| 448 |
+
"name": "knnbg_count",
|
| 449 |
+
"gain": 20.121904373168945
|
| 450 |
+
},
|
| 451 |
+
{
|
| 452 |
+
"name": "knnmp_log_min",
|
| 453 |
+
"gain": 19.98278045654297
|
| 454 |
+
},
|
| 455 |
+
{
|
| 456 |
+
"name": "n_unique_chars",
|
| 457 |
+
"gain": 19.560546875
|
| 458 |
+
},
|
| 459 |
+
{
|
| 460 |
+
"name": "kind__random",
|
| 461 |
+
"gain": 19.375856399536133
|
| 462 |
+
},
|
| 463 |
+
{
|
| 464 |
+
"name": "starts_digit",
|
| 465 |
+
"gain": 18.9150390625
|
| 466 |
+
},
|
| 467 |
+
{
|
| 468 |
+
"name": "origin__chinese",
|
| 469 |
+
"gain": 18.488323211669922
|
| 470 |
+
},
|
| 471 |
+
{
|
| 472 |
+
"name": "eth_tvl_usd",
|
| 473 |
+
"gain": 18.1763858795166
|
| 474 |
+
},
|
| 475 |
+
{
|
| 476 |
+
"name": "n_lower",
|
| 477 |
+
"gain": 17.70460319519043
|
| 478 |
+
},
|
| 479 |
+
{
|
| 480 |
+
"name": "n_letters",
|
| 481 |
+
"gain": 17.535446166992188
|
| 482 |
+
},
|
| 483 |
+
{
|
| 484 |
+
"name": "pca_004",
|
| 485 |
+
"gain": 16.79183578491211
|
| 486 |
+
},
|
| 487 |
+
{
|
| 488 |
+
"name": "kind__abbreviation",
|
| 489 |
+
"gain": 16.659326553344727
|
| 490 |
+
},
|
| 491 |
+
{
|
| 492 |
+
"name": "desc_pca_000",
|
| 493 |
+
"gain": 16.628366470336914
|
| 494 |
+
},
|
| 495 |
+
{
|
| 496 |
+
"name": "is_palindrome",
|
| 497 |
+
"gain": 16.464502334594727
|
| 498 |
+
},
|
| 499 |
+
{
|
| 500 |
+
"name": "fg_value",
|
| 501 |
+
"gain": 15.839948654174805
|
| 502 |
+
}
|
| 503 |
+
],
|
| 504 |
+
"family_gain_split": {
|
| 505 |
+
"mpnet_ft_knn": 217.51888847351074,
|
| 506 |
+
"bge_knn": 67.69801044464111,
|
| 507 |
+
"llm_kind": 68.03838443756104,
|
| 508 |
+
"llm_origin": 118.48860311508179,
|
| 509 |
+
"llm_scores": 37.4162712097168,
|
| 510 |
+
"llm_desc": 208.6804609298706,
|
| 511 |
+
"llm_total": 432.6237196922302
|
| 512 |
+
},
|
| 513 |
+
"inference_recipe": {
|
| 514 |
+
"description": "Inference uses 3 models + calibration constants",
|
| 515 |
+
"point_estimate": "final_log = q50_model(features)",
|
| 516 |
+
"uncalibrated_low": "low_log_raw = q05_model(features)",
|
| 517 |
+
"uncalibrated_high": "high_log_raw = q95_model(features)",
|
| 518 |
+
"calibrated_low_additive": "low_log = low_log_raw - delta_lower_additive",
|
| 519 |
+
"calibrated_high_additive": "high_log = high_log_raw + delta_upper_additive",
|
| 520 |
+
"calibrated_low_multiplicative": "low_log = q50 - multiplicative_factor * (q50 - low_log_raw)",
|
| 521 |
+
"calibrated_high_multiplicative": "high_log = q50 + multiplicative_factor * (high_log_raw - q50)",
|
| 522 |
+
"use_method": "additive",
|
| 523 |
+
"output_usd": "np.exp(final_log)"
|
| 524 |
+
},
|
| 525 |
+
"wandb_run": "https://wandb.ai/quantumly-aletheia-research/ens-appraiser/runs/xd2jhbwk"
|
| 526 |
+
}
|