Level-Adaptive Replay Memory / bench_report.json

Failed on benchmark

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "tabular",
  4  "model": "mlp_tiny",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.01,
 10      "depth": 8
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "depth": 1
 17        },
 18        "mean": 217.19686126708984
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.001,
 23          "depth": 2
 24        },
 25        "mean": 179.02721405029297
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.001,
 30          "depth": 4
 31        },
 32        "mean": 25.813298225402832
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.001,
 37          "depth": 8
 38        },
 39        "mean": 15.061450004577637
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.003,
 44          "depth": 1
 45        },
 46        "mean": 142.7078514099121
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.003,
 51          "depth": 2
 52        },
 53        "mean": 26.81587314605713
 54      },
 55      {
 56        "cfg": {
 57          "lr": 0.003,
 58          "depth": 4
 59        },
 60        "mean": 13.492629766464233
 61      },
 62      {
 63        "cfg": {
 64          "lr": 0.003,
 65          "depth": 8
 66        },
 67        "mean": 9.146221399307251
 68      },
 69      {
 70        "cfg": {
 71          "lr": 0.01,
 72          "depth": 1
 73        },
 74        "mean": 35.42466497421265
 75      },
 76      {
 77        "cfg": {
 78          "lr": 0.01,
 79          "depth": 2
 80        },
 81        "mean": 16.24430751800537
 82      },
 83      {
 84        "cfg": {
 85          "lr": 0.01,
 86          "depth": 4
 87        },
 88        "mean": 8.620880007743835
 89      },
 90      {
 91        "cfg": {
 92          "lr": 0.01,
 93          "depth": 8
 94        },
 95        "mean": 7.093326091766357
 96      }
 97    ],
 98    "full": {
 99      "mean": 7.334557235240936,
100      "std": 0.824177040857451,
101      "per_seed": [
102        6.754525661468506,
103        5.7614359855651855,
104        7.475412368774414,
105        8.381930351257324,
106        7.79284143447876,
107        6.628401756286621,
108        8.117091178894043,
109        7.764819145202637
110      ],
111      "n": 8
112    }
113  },
114  "idea": {
115    "mean": 23.1199893951416,
116    "std": 5.121285566952831,
117    "per_seed": [
118      15.503646850585938,
119      19.21010398864746,
120      18.322921752929688,
121      20.51085662841797,
122      24.305091857910156,
123      29.11318016052246,
124      29.386009216308594,
125      28.608104705810547
126    ],
127    "n": 8,
128    "selected_cfg": {
129      "lr": 0.01
130    }
131  },
132  "comparison": {
133    "delta_mean": 15.785432159900665,
134    "idea_wins": 0,
135    "n_pairs": 8,
136    "per_seed_diffs": [
137      8.749121189117432,
138      13.448668003082275,
139      10.847509384155273,
140      12.128926277160645,
141      16.512250423431396,
142      22.48477840423584,
143      21.26891803741455,
144      20.84328556060791
145    ],
146    "p_value": 0.0081,
147    "mde": 4.399418320126617,
148    "mde_rel_pct": 59.982057253413714,
149    "verdict": "idea worse (significant)",
150    "system_worked": false
151  },
152  "mechanism_signature": {
153    "math_sanity": {
154      "formula_check": [
155        {
156          "M": 1,
157          "predicted": 0.17300000000000004,
158          "observed": 0.17465833333333333,
159          "abs_error": 0.0016583333333332895
160        },
161        {
162          "M": 2,
163          "predicted": 0.3160710000000001,
164          "observed": 0.3166833333333333,
165          "abs_error": 0.0006123333333332148
166        },
167        {
168          "M": 4,
169          "predicted": 0.5322411229590001,
170          "observed": 0.5315916666666667,
171          "abs_error": 0.0006494562923333724
172        },
173        {
174          "M": 8,
175          "predicted": 0.7812016329493428,
176          "observed": 0.7828416666666667,
177          "abs_error": 0.0016400337173239077
178        },
179        {
180          "M": 16,
181          "predicted": 0.9521272745759659,
182          "observed": 0.9519,
183          "abs_error": 0.00022727457596594114
184        },
185        {
186          "M": 32,
187          "predicted": 0.9977082021604751,
188          "observed": 0.997875,
189          "abs_error": 0.00016679783952489036
190        }
191      ],
192      "max_abs_error": 0.0016583333333332895,
193      "stale_curve_optima": [
194        {
195          "stale": 0.0,
196          "optimal_depth": 16
197        },
198        {
199          "stale": 0.005,
200          "optimal_depth": 7
201        },
202        {
203          "stale": 0.015,
204          "optimal_depth": 4
205        },
206        {
207          "stale": 0.03,
208          "optimal_depth": 3
209        },
210        {
211          "stale": 0.05,
212          "optimal_depth": 3
213        }
214      ]
215    },
216    "trained_model_signature": {
217      "rho_hat": 0.13499999046325684,
218      "predicted_escape": 0.44015932468572094,
219      "observed_group_escape": 0.4399999976158142,
220      "absolute_error": 0.00015932706990673395,
221      "confirmed": true,
222      "mean_depth": 1.0,
223      "depth_trajectory": [
224        1,
225        1,
226        1,
227        1,
228        1,
229        1,
230        1,
231        1,
232        1,
233        1,
234        1,
235        1,
236        1,
237        1,
238        1,
239        1,
240        1,
241        1
242      ]
243    },
244    "idea_lr_trials": [
245      {
246        "cfg": {
247          "lr": 0.01
248        },
249        "result": {
250          "mean": 23.1199893951416,
251          "std": 5.121285566952831,
252          "per_seed": [
253            15.503646850585938,
254            19.21010398864746,
255            18.322921752929688,
256            20.51085662841797,
257            24.305091857910156,
258            29.11318016052246,
259            29.386009216308594,
260            28.608104705810547
261          ],
262          "n": 8,
263          "selected_cfg": {
264            "lr": 0.01
265          }
266        }
267      },
268      {
269        "cfg": {
270          "lr": 0.001
271        },
272        "result": {
273          "mean": 215.88136672973633,
274          "std": 10.926122378078592,
275          "per_seed": [
276            228.2001190185547,
277            218.22589111328125,
278            219.18821716308594,
279            203.1732177734375,
280            216.90838623046875,
281            202.43487548828125,
282            205.00531005859375,
283            233.9149169921875
284          ],
285          "n": 8
286        }
287      },
288      {
289        "cfg": {
290          "lr": 0.003
291        },
292        "result": {
293          "mean": 140.52062129974365,
294          "std": 14.768009273435721,
295          "per_seed": [
296            146.7877197265625,
297            148.9251708984375,
298            133.30699157714844,
299            141.8115234375,
300            114.57950592041016,
301            140.2008819580078,
302            129.85032653808594,
303            168.70285034179688
304          ],
305          "n": 8
306        }
307      }
308    ],
309    "track_choice": "tabular: replay-memory training intervention; same MLP and regression task in both arms"
310  }
311}