fix(examples): CAVIAR deep-review follow-ups — vocabulary reproducibility, shipped audit artifacts, canonical close_34

.github/workflows/ci.yml

@@ -95,6 +95,37 @@ jobs:
       - name: Validate Python provenance stubs and documentation
         run: python -m pytest -q python/tests/test_relation_provenance_contract.py
 
+  # The CAVIAR example suites on CPU (review follow-up: previously no
+  # workflow ran any of them). CPU torch makes the audit/census/corpus
+  # suites run for real instead of module-skipping; the dataset-gated
+  # tests inside them self-skip (no corpus data ships in the repo), and
+  # the GPU-only suites stay local per the hardware table in
+  # docs/experiments/caviar/README.md.
+  caviar-examples:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6
+
+      - uses: actions/setup-python@a309ff8b426b58ec0e2a45f0f869d46889d02405 # v6
+        with:
+          python-version: "3.11"
+
+      - name: Install pytest and CPU torch
+        run: |
+          python -m pip install --upgrade pip pytest
+          python -m pip install torch --index-url https://download.pytorch.org/whl/cpu
+
+      - name: Run CAVIAR example suites
+        env:
+          PYTHONPATH: crates/pyxlog/python
+        run: >-
+          python -m pytest -q
+          python/tests/test_theory_loop.py
+          python/tests/test_caviar_xml_corpus.py
+          python/tests/test_audit_dump_vs_xml.py
+          python/tests/test_xml_meeting_census.py
+          python/tests/test_audit_tool_torch_free.py
+
   fmt:
     runs-on: ubuntu-latest
     steps:

docs/experiments/caviar/results/e9_f1_holdout/caviar-e9-cont_rel_ec_mnc2_f1holdout_fixed075.json

@@ -0,0 +1,236 @@
+{
+  "pkl": "C:/Users/User/flt-workspace/projects/xlog/runs/r_djwq4hnn0dqoi4e2/caviar_data_continuous/caviar-data/caviar-train.json",
+  "test_json": "C:/Users/User/flt-workspace/projects/xlog/runs/r_djwq4hnn0dqoi4e2/caviar_data_continuous/caviar-data/caviar-test.json",
+  "data": "continuous",
+  "fold": null,
+  "mode": "relational",
+  "protocol": "ec",
+  "close_threshold": 25.0,
+  "k": 4,
+  "seed": 7,
+  "hidden": 16,
+  "max_clauses": 4,
+  "min_new_covered": 2,
+  "max_body_literals": 3,
+  "holdout_score": "f1",
+  "transition_vocab": "activity",
+  "ec_fit_mode": "fixed",
+  "min_fit": null,
+  "null_permutations": 1000,
+  "null_quantile": 0.95,
+  "null_perm_seed": 7,
+  "tie_tolerance": null,
+  "num_pt": {
+    "train": 24312,
+    "test": 8048
+  },
+  "n_pos": {
+    "train": 1344,
+    "test": 489
+  },
+  "wall_clock_s": {
+    "convert": 0.6520058000460267,
+    "theory_loop_init": 0.38313480000942945,
+    "theory_loop_term": 0.12887379992753267,
+    "mode_total": 0.5593090999173,
+    "total": 1.2967943999683484
+  },
+  "candidate_vocabulary": {
+    "relational": [
+      "any_became_active",
+      "any_became_inactive",
+      "any_became_walking",
+      "any_stopped_walking",
+      "both_active",
+      "both_inactive",
+      "both_walking",
+      "close",
+      "far",
+      "mixed_active_walking"
+    ],
+    "neural": [],
+    "excluded": [
+      "coords_missing"
+    ]
+  },
+  "steps_requested": null,
+  "steps_effective": null,
+  "steps_clamped": false,
+  "ec": {
+    "n_init": {
+      "train": 10,
+      "test": 3
+    },
+    "n_term": {
+      "train": 11,
+      "test": 1
+    },
+    "n_init_dontcare_train": 1334,
+    "n_term_dontcare_train": 22957,
+    "init_theory": {
+      "clauses": [],
+      "iterations": [
+        {
+          "rule": null,
+          "reason": "select_once abstained",
+          "margin": 0.0,
+          "n_residual_pos_before": 10,
+          "n_newly_covered": 0
+        }
+      ],
+      "stop_reason": "select_once abstained"
+    },
+    "term_theory": {
+      "clauses": [],
+      "iterations": [
+        {
+          "rule": null,
+          "reason": "select_once abstained",
+          "margin": 0.0,
+          "n_residual_pos_before": 11,
+          "n_newly_covered": 0
+        }
+      ],
+      "stop_reason": "select_once abstained"
+    },
+    "init_scoring": {
+      "theory_prf1": {
+        "train": {
+          "precision": 0.0,
+          "recall": 0.0,
+          "f1": 0.0,
+          "tp": 0,
+          "fp": 0,
+          "fn": 10,
+          "tn": 24302,
+          "degenerate": true
+        },
+        "test": {
+          "precision": 0.0,
+          "recall": 0.0,
+          "f1": 0.0,
+          "tp": 0,
+          "fp": 0,
+          "fn": 3,
+          "tn": 8045,
+          "degenerate": true
+        }
+      },
+      "marginal_contribution": []
+    },
+    "term_scoring": {
+      "theory_prf1": {
+        "train": {
+          "precision": 0.0,
+          "recall": 0.0,
+          "f1": 0.0,
+          "tp": 0,
+          "fp": 0,
+          "fn": 11,
+          "tn": 24301,
+          "degenerate": true
+        },
+        "test": {
+          "precision": 0.0,
+          "recall": 0.0,
+          "f1": 0.0,
+          "tp": 0,
+          "fp": 0,
+          "fn": 1,
+          "tn": 8047,
+          "degenerate": true
+        }
+      },
+      "marginal_contribution": []
+    },
+    "frame_f1": {
+      "precision": 0.0,
+      "recall": 0.0,
+      "f1": 0.0,
+      "tp": 0,
+      "fp": 0,
+      "fn": 489,
+      "tn": 7559,
+      "degenerate": true
+    },
+    "relational_search_pool": {
+      "init": {
+        "bodies_by_size": {
+          "2": 30,
+          "3": 29
+        },
+        "skipped_empty_cover": {
+          "2": 15,
+          "3": 91
+        }
+      },
+      "term": {
+        "bodies_by_size": {
+          "2": 30,
+          "3": 29
+        },
+        "skipped_empty_cover": {
+          "2": 15,
+          "3": 91
+        }
+      }
+    },
+    "init_scores_last_iteration_top5": [
+      [
+        "both_active&close",
+        0.23809523809523808
+      ],
+      [
+        "any_became_active&both_active&close",
+        0.1
+      ],
+      [
+        "any_became_active&both_active",
+        0.08333333333333333
+      ],
+      [
+        "any_became_active&any_stopped_walking&mixed_active_walking",
+        0.07142857142857144
+      ],
+      [
+        "any_became_active&close",
+        0.07142857142857144
+      ]
+    ],
+    "term_scores_last_iteration_top5": [
+      [
+        "any_became_inactive&any_became_walking",
+        0.1
+      ],
+      [
+        "any_became_inactive&any_became_walking&far",
+        0.1
+      ],
+      [
+        "any_became_inactive&far",
+        0.1
+      ],
+      [
+        "any_became_walking&far",
+        0.1
+      ],
+      [
+        "both_walking&close",
+        0.1
+      ]
+    ],
+    "init_selection_reasons": [
+      "no body passed the fit gate (min_fit=0.75): a body that cannot fit held-out data is not a rule"
+    ],
+    "term_selection_reasons": [
+      "no body passed the fit gate (min_fit=0.75): a body that cannot fit held-out data is not a rule"
+    ],
+    "ec_fit_mode": "fixed",
+    "init_min_fit": 0.75,
+    "term_min_fit": 0.75,
+    "init_null_summary": null,
+    "term_null_summary": null
+  },
+  "direct_context": null,
+  "detector_probe": null
+}