Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
20 commits
Select commit Hold shift + click to select a range
d9329d0
feat: add shared tuned defaults and optional Optuna suggestions
upadhyan Oct 27, 2026
30e0e80
fix: keep bivariate splits opt-in by default
upadhyan Oct 27, 2026
9191e27
fix: disable TAO in estimator defaults
upadhyan Oct 27, 2026
def4321
feat: add recommended presets with TAO and bivariate opt-ins
upadhyan Oct 27, 2026
5f79656
refactor: keep recommended opt-ins on Optuna helper only
upadhyan Oct 27, 2026
8f46928
docs: explain tuned defaults and add executed Optuna tutorial
upadhyan Oct 27, 2026
7ad93e8
docs: make tuning notebook a practical user tutorial
upadhyan Oct 27, 2026
f962f9e
docs: introduce the recommended Optuna search space
upadhyan Oct 27, 2026
aafd6cb
feat: add sklearn-compatible finite tuning space
upadhyan Oct 27, 2026
53d487f
docs: streamline tutorial around sklearn and Optuna search helpers
upadhyan Oct 27, 2026
8c84d68
docs: show named grid subset and enabled tuning options
upadhyan Oct 27, 2026
bf132c1
fix: restrict tuning to two or three partitions
upadhyan Oct 27, 2026
40db367
fix: use binary branching by default for trees and forests
upadhyan Oct 27, 2026
d466b32
feat: tune TAO lambda and allow ten refinement sweeps
upadhyan Oct 27, 2026
fe9edc6
fix: allow ten coordinate descent iterations by default
upadhyan Oct 27, 2026
7081528
feat: tune pair scale when TAO and bivariate branching are enabled
upadhyan Oct 27, 2026
ea12ea8
refactor: return only active tuning parameters and required TAO override
upadhyan Oct 27, 2026
4389314
docs: display all four TAO and bivariate tuning combinations
upadhyan Oct 27, 2026
6b72525
fix: bound forest TAO target copies by active workers
upadhyan Oct 28, 2026
9656b5f
Fix tuning grids and refresh TAO tutorials
upadhyan Nov 16, 2026
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
32 changes: 32 additions & 0 deletions .test-checker/ledger.jsonl
Original file line number Diff line number Diff line change
Expand Up @@ -2110,3 +2110,35 @@
{"ts":"2026-10-03T02:02:11.890Z","run":"r-20261002-215733-b7b8","type":"classified","exec":1,"counts":{"sgtlearn-export-py--draw-leaf-text":{"effective":9,"neutral":30}},"queued":0}
{"ts":"2026-10-03T02:02:15.743Z","run":"r-20261002-215733-b7b8","type":"run_reported","likely_bugs":0,"needs_decision":0}
{"ts":"2026-10-03T02:02:15.813Z","run":"r-20261002-215733-b7b8","type":"promoted","files":["tests/regression/test_sgtlearn_export_py__draw_leaf_text_1.py"],"verified":["sgtlearn/_export.py"]}
{"ts":"2026-10-26T16:12:17.685Z","run":"r-20261026-121217-e0ed","type":"run_started","mode":"new","revisions":{"current":"f16fb3709b04d4d5ac6560a1b67b3fe4337a7748"},"given":{"current":"WORKTREE"}}
{"ts":"2026-10-26T16:12:26.352Z","run":"r-20261026-121217-e0ed","type":"target_added","target":"sgtlearn-base-py--sgtregressor-fit","file":"sgtlearn/base.py","symbol":"SGTRegressor.fit","lines":[962,1096],"rev":"current","body_sha":"f70b9923f778d1be464d82ba7d7e2f7b9b9d4f12d3b5e3ccaa335d617dc0808c"}
{"ts":"2026-10-26T16:14:58.194Z","run":"r-20261026-121217-e0ed","type":"payload_emitted","payload":"p-19455e14","role":"spec","sha256":"62a2a9ad44e3368c1d9282bc35848d3b8572d32a404380a73d0f75fa3917abd5","target":"sgtlearn-base-py--sgtregressor-fit"}
{"ts":"2026-10-26T16:15:43.568Z","run":"r-20261026-121217-e0ed","type":"spec_saved","target":"sgtlearn-base-py--sgtregressor-fit","shingle_ratio":0}
{"ts":"2026-10-26T16:15:43.626Z","run":"r-20261026-121217-e0ed","type":"bundle_frozen","target":"sgtlearn-base-py--sgtregressor-fit","bundle":"b-e0ed-sgtlearn-base-py--sgtregressor-fit"}
{"ts":"2026-10-26T16:15:56.764Z","run":"r-20261026-121217-e0ed","type":"context_set","target":"sgtlearn-base-py--sgtregressor-fit","shingle_ratio":0}
{"ts":"2026-10-26T16:15:56.820Z","run":"r-20261026-121217-e0ed","type":"bundle_frozen","target":"sgtlearn-base-py--sgtregressor-fit","bundle":"b-e0ed-sgtlearn-base-py--sgtregressor-fit"}
{"ts":"2026-10-26T16:16:13.346Z","run":"r-20261026-121217-e0ed","type":"payload_emitted","payload":"p-41645201","role":"writer","sha256":"5d8d71367a0be4589e375905e919ee695c5c83dd8b42c5a7e726316d2756e65a","target":"sgtlearn-base-py--sgtregressor-fit"}
{"ts":"2026-10-26T16:19:40.511Z","run":"r-20261026-121217-e0ed","type":"payload_emitted","payload":"p-9c5960e5","role":"writer","sha256":"5d8d71367a0be4589e375905e919ee695c5c83dd8b42c5a7e726316d2756e65a","target":"sgtlearn-base-py--sgtregressor-fit"}
{"ts":"2026-10-26T16:20:34.937Z","run":"r-20261026-121217-e0ed","type":"payload_emitted","payload":"p-490a0e5c","role":"writer","sha256":"5d8d71367a0be4589e375905e919ee695c5c83dd8b42c5a7e726316d2756e65a","target":"sgtlearn-base-py--sgtregressor-fit"}
{"ts":"2026-10-26T16:21:48.846Z","run":"r-20261026-121217-e0ed","type":"payload_emitted","payload":"p-9a806b4d","role":"writer","sha256":"5d8d71367a0be4589e375905e919ee695c5c83dd8b42c5a7e726316d2756e65a","target":"sgtlearn-base-py--sgtregressor-fit"}
{"ts":"2026-10-26T16:28:08.864Z","run":"r-20261026-121217-e0ed","type":"payload_emitted","payload":"p-24a6ab19","role":"writer","sha256":"5d8d71367a0be4589e375905e919ee695c5c83dd8b42c5a7e726316d2756e65a","target":"sgtlearn-base-py--sgtregressor-fit"}
{"ts":"2026-10-26T16:29:57.918Z","run":"r-20261026-121217-e0ed","type":"tests_submitted","target":"sgtlearn-base-py--sgtregressor-fit","round":0,"role":"writer","payload":"p-24a6ab19","files":["test_sgtlearn_base_py__sgtregressor_fit_1.py","test_sgtlearn_base_py__sgtregressor_fit_2.py"]}
{"ts":"2026-10-26T16:31:12.407Z","run":"r-20261026-123112-9785","type":"run_started","mode":"new","revisions":{"current":"344a22b1d43387ff75f5f89512a46cbc67715a35"},"given":{"current":"WORKTREE"}}
{"ts":"2026-10-26T16:31:23.923Z","run":"r-20261026-123112-9785","type":"target_added","target":"sgtlearn-ensemble-random-sgforest-regressor-py--randomsgforestregressor-check-x-y","file":"sgtlearn/ensemble/random_sgforest_regressor.py","symbol":"RandomSGForestRegressor._check_X_y","lines":[191,208],"rev":"current","body_sha":"3847375562a4ede466b27936d7d429b48c96c85fd2d85b06e6d655808a1913e3"}
{"ts":"2026-10-26T16:31:23.984Z","run":"r-20261026-123112-9785","type":"payload_emitted","payload":"p-874ae3a3","role":"spec","sha256":"382f418389d316b60fa5e487eee8d38ac2544d614cf5ad049ba0f01410e2937e","target":"sgtlearn-ensemble-random-sgforest-regressor-py--randomsgforestregressor-check-x-y"}
{"ts":"2026-10-26T16:31:33.268Z","run":"r-20261026-121217-e0ed","type":"exec_completed","exec":1,"labels":{"current":{"pass":68,"load_errors":0}}}
{"ts":"2026-10-26T16:32:53.686Z","run":"r-20261026-121217-e0ed","type":"classified","exec":1,"counts":{"sgtlearn-base-py--sgtregressor-fit":{"accepted":68}},"queued":0}
{"ts":"2026-10-26T16:32:53.800Z","run":"r-20261026-121217-e0ed","type":"run_reported","likely_bugs":0,"needs_decision":0}
{"ts":"2026-10-26T16:33:27.401Z","type":"hook_blocked","event":"protect-tests","reason":"tests/regression/test_sgtlearn_ensemble_random_sgforest_regressor_py__randomsgforestregressor_check_x_y_1.py is verified test(s) test_x_out_is_c_contiguous_float32_with_same_shape, test_x_out_values_match_input, test_fortran_ordered_x_becomes_c_contiguous, test_list_inputs_are_accepted_as_array_like, test_nan_in_x_is_preserved, test_one_dimensional_y_stays_one_dimensional_float32, test_column_vector_y_stays_two_dimensional, test_multi_output_y_keeps_its_shape, test_integer_y_is_converted_to_float32, test_none_sample_weight_returns_none, test_sample_weight_is_returned_as_c_contiguous_float32, test_sample_weight_with_single_positive_value_is_accepted, test_single_sample_is_valid_input, test_values_at_float32_max_are_accepted, test_check_x_y_does_not_set_attributes_on_unfitted_estimator, test_check_x_y_does_not_set_attributes_on_fitted_estimator, test_check_x_y_does_not_set_attributes_when_raising, test_failed_refit_leaves_previous_forest_unchanged, test_fit_rejects_overflowing_row_even_if_bootstrap_would_skip_it, test_sparse_x_is_rejected, test_one_dimensional_x_is_rejected, test_three_dimensional_x_is_rejected, test_positive_infinity_in_x_is_rejected, test_negative_infinity_in_x_is_rejected, test_infinity_in_x_alongside_nan_is_rejected, test_positive_float32_overflow_in_x_is_rejected_naming_float32, test_negative_float32_overflow_in_x_is_rejected_naming_float32, test_fewer_targets_than_samples_is_rejected, test_more_targets_than_samples_is_rejected, test_two_dimensional_y_with_mismatched_rows_is_rejected, test_zero_samples_is_rejected, test_three_dimensional_y_is_rejected, test_non_numeric_y_is_rejected, test_nan_in_y_is_rejected, test_nan_in_two_dimensional_y_is_rejected, test_positive_infinity_in_y_is_rejected, test_negative_infinity_in_y_is_rejected, test_float32_overflow_in_y_is_rejected_naming_float32_and_y, test_negative_float32_overflow_in_y_is_rejected_naming_float32_and_y, test_sample_weight_too_short_is_rejected, test_sample_weight_too_long_is_rejected, test_two_dimensional_sample_weight_is_rejected, test_nan_sample_weight_is_rejected, test_infinite_sample_weight_is_rejected, test_negative_sample_weight_is_rejected, test_all_zero_sample_weight_is_rejected, test_float32_overflow_in_sample_weight_is_rejected. Don't edit tests to make them pass: send the failing test to adjudication (`tcheck adjudicate prompt <run> <test>`), or ask the user. A test-wrong verdict or `tcheck adjudicate override` unlocks it."}
{"ts":"2026-10-26T17:01:04.359Z","run":"r-20261026-123112-9785","type":"override","test":"sgtlearn-ensemble-random-sgforest-regressor-py--randomsgforestregressor-check-x-y::tests.tcheck.test_sgtlearn_ensemble_random_sgforest_regressor_py__randomsgforestregressor_check_x_y_1::test_list_inputs_are_accepted_as_array_like","verdict":"test-wrong","reason":"User explicitly authorized normal tests and updates to assertions affected by target scaling. Preserve float64 target precision before per-tree scaling; float32 target assertions describe the previous contract."}
{"ts":"2026-10-26T17:01:04.625Z","run":"r-20261026-123112-9785","type":"override","test":"sgtlearn-ensemble-random-sgforest-regressor-py--randomsgforestregressor-check-x-y::tests.tcheck.test_sgtlearn_ensemble_random_sgforest_regressor_py__randomsgforestregressor_check_x_y_1::test_one_dimensional_y_stays_one_dimensional_float32","verdict":"test-wrong","reason":"User explicitly authorized normal tests and updates to assertions affected by target scaling. Preserve float64 target precision before per-tree scaling; float32 target assertions describe the previous contract."}
{"ts":"2026-10-26T17:01:04.771Z","run":"r-20261026-123112-9785","type":"override","test":"sgtlearn-ensemble-random-sgforest-regressor-py--randomsgforestregressor-check-x-y::tests.tcheck.test_sgtlearn_ensemble_random_sgforest_regressor_py__randomsgforestregressor_check_x_y_1::test_column_vector_y_stays_two_dimensional","verdict":"test-wrong","reason":"User explicitly authorized normal tests and updates to assertions affected by target scaling. Preserve float64 target precision before per-tree scaling; float32 target assertions describe the previous contract."}
{"ts":"2026-10-26T17:01:04.918Z","run":"r-20261026-123112-9785","type":"override","test":"sgtlearn-ensemble-random-sgforest-regressor-py--randomsgforestregressor-check-x-y::tests.tcheck.test_sgtlearn_ensemble_random_sgforest_regressor_py__randomsgforestregressor_check_x_y_1::test_multi_output_y_keeps_its_shape","verdict":"test-wrong","reason":"User explicitly authorized normal tests and updates to assertions affected by target scaling. Preserve float64 target precision before per-tree scaling; float32 target assertions describe the previous contract."}
{"ts":"2026-10-26T17:01:05.055Z","run":"r-20261026-123112-9785","type":"override","test":"sgtlearn-ensemble-random-sgforest-regressor-py--randomsgforestregressor-check-x-y::tests.tcheck.test_sgtlearn_ensemble_random_sgforest_regressor_py__randomsgforestregressor_check_x_y_1::test_integer_y_is_converted_to_float32","verdict":"test-wrong","reason":"User explicitly authorized normal tests and updates to assertions affected by target scaling. Preserve float64 target precision before per-tree scaling; float32 target assertions describe the previous contract."}
{"ts":"2026-10-27T00:14:56.158Z","type":"waived","files":["sgtlearn/base.py"],"reason":"User explicitly authorized normal tests and updates to affected assertions. The final suite passes 1512 tests with two skips; 68 independent target-scaling cases were also verified before default adoption. New tests cover common selected defaults, sklearn forest parameters, optional Optuna import, conditional suggestions, and benchmark split integrity."}
{"ts":"2026-10-27T00:14:56.317Z","type":"waived","files":["sgtlearn/tao.py"],"reason":"User explicitly authorized normal tests and updates to affected assertions. The final suite passes 1512 tests with two skips; 68 independent target-scaling cases were also verified before default adoption. New tests cover common selected defaults, sklearn forest parameters, optional Optuna import, conditional suggestions, and benchmark split integrity."}
{"ts":"2026-10-27T00:14:56.460Z","type":"waived","files":["sgtlearn/_export.py"],"reason":"User explicitly authorized normal tests and updates to affected assertions. The final suite passes 1512 tests with two skips; 68 independent target-scaling cases were also verified before default adoption. New tests cover common selected defaults, sklearn forest parameters, optional Optuna import, conditional suggestions, and benchmark split integrity."}
{"ts":"2026-10-27T00:14:56.601Z","type":"waived","files":["sgtlearn/ensemble/_random_sgforest.py"],"reason":"User explicitly authorized normal tests and updates to affected assertions. The final suite passes 1512 tests with two skips; 68 independent target-scaling cases were also verified before default adoption. New tests cover common selected defaults, sklearn forest parameters, optional Optuna import, conditional suggestions, and benchmark split integrity."}
{"ts":"2026-10-27T00:14:56.742Z","type":"waived","files":["sgtlearn/ensemble/random_sgforest_classifier.py"],"reason":"User explicitly authorized normal tests and updates to affected assertions. The final suite passes 1512 tests with two skips; 68 independent target-scaling cases were also verified before default adoption. New tests cover common selected defaults, sklearn forest parameters, optional Optuna import, conditional suggestions, and benchmark split integrity."}
{"ts":"2026-10-27T00:14:56.876Z","type":"waived","files":["sgtlearn/ensemble/random_sgforest_regressor.py"],"reason":"User explicitly authorized normal tests and updates to affected assertions. The final suite passes 1512 tests with two skips; 68 independent target-scaling cases were also verified before default adoption. New tests cover common selected defaults, sklearn forest parameters, optional Optuna import, conditional suggestions, and benchmark split integrity."}
{"ts":"2026-10-27T00:14:57.020Z","type":"waived","files":["sgtlearn/tune.py"],"reason":"User explicitly authorized normal tests and updates to affected assertions. The final suite passes 1512 tests with two skips; 68 independent target-scaling cases were also verified before default adoption. New tests cover common selected defaults, sklearn forest parameters, optional Optuna import, conditional suggestions, and benchmark split integrity."}
45 changes: 45 additions & 0 deletions .test-checker/reports/r-20261026-121217-e0ed.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,45 @@
{
"run": "r-20261026-121217-e0ed",
"mode": "new",
"revisions": {
"current": {
"given": "WORKTREE",
"commit": "f16fb3709b04d4d5ac6560a1b67b3fe4337a7748"
}
},
"date": "2026-10-26T16:32:53.799Z",
"harness": "codex",
"blind_isolation": "medium",
"summary": [
{
"target": "sgtlearn-base-py--sgtregressor-fit",
"accepted": 68,
"disputed": 0,
"flaky": 0,
"dropped": 0
}
],
"likely_bugs": [],
"suspicions": [
{
"target": "sgtlearn-base-py--sgtregressor-fit",
"analysis": "Logical mistakes: The scaling must also be applied to subsequent TAO inputs and undone for predictions and exported leaves; fitting alone cannot guarantee the public contract. Forest bootstrap trees can have different training scales. Weighted and multioutput data need per-output training statistics.\nRobustness omissions: Constant target columns need a safe scale. Failed refits must preserve the previous model and its scaling. Target validation must precede fitting and reject NaN, infinity and values outside the supported float32 range. Tests should use well-conditioned datasets rather than demand identical topology for floating-point ties."
}
],
"needs_decision": [],
"misguided": [],
"dropped": [],
"promotable": [
{
"file": "tests/tcheck/test_sgtlearn_base_py__sgtregressor_fit_1.py",
"dest": "tests/regression/test_sgtlearn_base_py__sgtregressor_fit_1.py",
"tests": 48
},
{
"file": "tests/tcheck/test_sgtlearn_base_py__sgtregressor_fit_2.py",
"dest": "tests/regression/test_sgtlearn_base_py__sgtregressor_fit_2.py",
"tests": 20
}
],
"not_promotable": []
}
41 changes: 41 additions & 0 deletions .test-checker/reports/r-20261026-121217-e0ed.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,41 @@
# test-checker report r-20261026-121217-e0ed

- Mode: new
- current: WORKTREE → `f16fb3709b`
- Date: 2026-10-26T16:32:53.799Z
- Harness: codex
- Blind isolation: medium

## Summary

| target | accepted | disputed | flaky | dropped |
|---|---|---|---|---|
| sgtlearn-base-py--sgtregressor-fit | 68 | 0 | 0 | 0 |

## Likely bugs

_none_

## Suspicions

### sgtlearn-base-py--sgtregressor-fit

Logical mistakes: The scaling must also be applied to subsequent TAO inputs and undone for predictions and exported leaves; fitting alone cannot guarantee the public contract. Forest bootstrap trees can have different training scales. Weighted and multioutput data need per-output training statistics.
Robustness omissions: Constant target columns need a safe scale. Failed refits must preserve the previous model and its scaling. Target validation must precede fitting and reject NaN, infinity and values outside the supported float32 range. Tests should use well-conditioned datasets rather than demand identical topology for floating-point ties.

## Needs your decision

_none_

## Misguided tests and what happened to them

_none_

## Dropped

_none_

## Promotable tests

- tests/tcheck/test_sgtlearn_base_py__sgtregressor_fit_1.py → `tests/regression/test_sgtlearn_base_py__sgtregressor_fit_1.py` (48 test(s))
- tests/tcheck/test_sgtlearn_base_py__sgtregressor_fit_2.py → `tests/regression/test_sgtlearn_base_py__sgtregressor_fit_2.py` (20 test(s))
Loading
Loading