Monotone CDT autoencoder bottleneck / bench_report.json

Mechanism confirmed, baseline not beaten

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "translated_density_transport",
  4  "model": "shared_mlp",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.009
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.6033583730459213
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.06099952198565006
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.009
 27        },
 28        "mean": 0.005193391698412597
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.004602075146976858,
 33      "std": 0.0010999913497623144,
 34      "per_seed": [
 35        0.004353613592684269,
 36        0.00704985624179244,
 37        0.005014418624341488,
 38        0.0043556783348321915,
 39        0.004588612355291843,
 40        0.0034741132985800505,
 41        0.004819120746105909,
 42        0.003161187982186675
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 4.106930315494537,
 49    "std": 0.0030807169256581463,
 50    "per_seed": [
 51      4.1090593338012695,
 52      4.107220649719238,
 53      4.104761600494385,
 54      4.1017680168151855,
 55      4.108211040496826,
 56      4.103658676147461,
 57      4.111758232116699,
 58      4.109004974365234
 59    ],
 60    "n": 8,
 61    "best_cfg": {
 62      "lr": 0.009
 63    },
 64    "sweep": [
 65      {
 66        "cfg": {
 67          "lr": 0.001
 68        },
 69        "mean": 4.110945641994476,
 70        "std": 0.003565288253498607,
 71        "per_seed": [
 72          4.113897800445557,
 73          4.113440036773682,
 74          4.1050310134887695,
 75          4.1087565422058105,
 76          4.1114397048950195,
 77          4.106204509735107,
 78          4.114856719970703,
 79          4.113938808441162
 80        ],
 81        "n": 8
 82      },
 83      {
 84        "cfg": {
 85          "lr": 0.003
 86        },
 87        "mean": 4.108930468559265,
 88        "std": 0.002267987204709172,
 89        "per_seed": [
 90          4.110673427581787,
 91          4.107449531555176,
 92          4.107608795166016,
 93          4.105695724487305,
 94          4.110192775726318,
 95          4.106705188751221,
 96          4.112852096557617,
 97          4.110266208648682
 98        ],
 99        "n": 8
100      },
101      {
102        "cfg": {
103          "lr": 0.009
104        },
105        "mean": 4.106930315494537,
106        "std": 0.0030807169256581463,
107        "per_seed": [
108          4.1090593338012695,
109          4.107220649719238,
110          4.104761600494385,
111          4.1017680168151855,
112          4.108211040496826,
113          4.103658676147461,
114          4.111758232116699,
115          4.109004974365234
116        ],
117        "n": 8
118      }
119    ]
120  },
121  "comparison": {
122    "delta_mean": 4.1023282403475605,
123    "idea_wins": 0,
124    "n_pairs": 8,
125    "per_seed_diffs": [
126      4.104705720208585,
127      4.100170793477446,
128      4.099747181870043,
129      4.097412338480353,
130      4.103622428141534,
131      4.100184562848881,
132      4.106939111370593,
133      4.105843786383048
134    ],
135    "p_value": 0.0081,
136    "mde": 0.0028447563556861727,
137    "mde_rel_pct": 61.81464371686579,
138    "verdict": "idea worse (significant)",
139    "system_worked": false
140  },
141  "mechanism_signature": {
142    "trained_models": true,
143    "prediction": "translated states should produce monotone quantiles with approximately constant +0.08 displacement, nonnegative decoded densities, and unit mass",
144    "predicted_vs_observed": {
145      "expected_quantile_shift": 0.08,
146      "observed_mean_quantile_shift": 0.05113657424226403,
147      "observed_monotone_fraction": 1.0,
148      "observed_idea_negative_fraction": 0.0,
149      "observed_idea_mean_mass_proxy_error": 2.808868893033889e-08,
150      "observed_baseline_negative_fraction": 0.3139062523841858
151    },
152    "confirmed": true,
153    "paired_values": [
154      {
155        "seed": 0,
156        "baseline": 0.004353613592684269,
157        "idea": 4.1090593338012695
158      },
159      {
160        "seed": 1,
161        "baseline": 0.00704985624179244,
162        "idea": 4.107220649719238
163      },
164      {
165        "seed": 2,
166        "baseline": 0.005014418624341488,
167        "idea": 4.104761600494385
168      },
169      {
170        "seed": 3,
171        "baseline": 0.0043556783348321915,
172        "idea": 4.1017680168151855
173      },
174      {
175        "seed": 4,
176        "baseline": 0.004588612355291843,
177        "idea": 4.108211040496826
178      },
179      {
180        "seed": 5,
181        "baseline": 0.0034741132985800505,
182        "idea": 4.103658676147461
183      },
184      {
185        "seed": 6,
186        "baseline": 0.004819120746105909,
187        "idea": 4.111758232116699
188      },
189      {
190        "seed": 7,
191        "baseline": 0.003161187982186675,
192        "idea": 4.109004974365234
193      }
194    ],
195    "idea_sweep": [
196      {
197        "cfg": {
198          "lr": 0.001
199        },
200        "mean": 4.110945641994476,
201        "std": 0.003565288253498607,
202        "per_seed": [
203          4.113897800445557,
204          4.113440036773682,
205          4.1050310134887695,
206          4.1087565422058105,
207          4.1114397048950195,
208          4.106204509735107,
209          4.114856719970703,
210          4.113938808441162
211        ],
212        "n": 8
213      },
214      {
215        "cfg": {
216          "lr": 0.003
217        },
218        "mean": 4.108930468559265,
219        "std": 0.002267987204709172,
220        "per_seed": [
221          4.110673427581787,
222          4.107449531555176,
223          4.107608795166016,
224          4.105695724487305,
225          4.110192775726318,
226          4.106705188751221,
227          4.112852096557617,
228          4.110266208648682
229        ],
230        "n": 8
231      },
232      {
233        "cfg": {
234          "lr": 0.009
235        },
236        "mean": 4.106930315494537,
237        "std": 0.0030807169256581463,
238        "per_seed": [
239          4.1090593338012695,
240          4.107220649719238,
241          4.104761600494385,
242          4.1017680168151855,
243          4.108211040496826,
244          4.103658676147461,
245          4.111758232116699,
246          4.109004974365234
247        ],
248        "n": 8
249      }
250    ]
251  }
252}