baselines: audit native author runs

1 files changed, 33 insertions, 0 deletions

diff --git a/experiments/baseline_smoke.py b/experiments/baseline_smoke.py
index c656a2f..324e333 100644
--- a/experiments/baseline_smoke.py
+++ b/experiments/baseline_smoke.py
@@ -8,6 +8,8 @@ sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
 from sdil.data import onehot
 from sdil.local_baselines import FANet, PEPITANet, FFNet, EPNet
 from sdil import probes
+from experiments.import_native_baseline import (summarize_burst_history,
+                                                summarize_dualprop_hist)
 
 
 def check_fa_residual_transport():
@@ -133,9 +135,40 @@ def check_ep_energy_dynamics():
     print("EP free particles persist across presentations")
 
 
+def check_native_result_selection():
+    burst_rows = [
+        {"epoch": 0, "epoch/top1_error/test": 30.0,
+         "epoch/top1_error/val": 28.0, "_runtime": 10.0},
+        {"epoch": 1, "epoch/top1_error/test": 20.0,
+         "epoch/top1_error/val": 22.0, "_runtime": 20.0},
+        {"epoch": 2, "epoch/top1_error/test": 21.0,
+         "epoch/top1_error/val": 18.0, "_runtime": 30.0},
+    ]
+    burst = summarize_burst_history(burst_rows, 3)
+    assert burst["primary_final_test_accuracy_percent"] == 79.0
+    assert burst["validation_selected_epoch"] == 2
+    assert burst["validation_selected_test_accuracy_percent"] == 79.0
+    assert burst["test_selected_epoch"] == 1
+    assert burst["test_selected_test_accuracy_percent"] == 80.0
+
+    dual = summarize_dualprop_hist({
+        "train_loss": [2.0, 1.0], "train_accuracy": [30.0, 70.0],
+        "train_time": [4.0, 5.0], "val_loss": [1.5, 1.2],
+        "val_accuracy": [60.0, 65.0], "val_top5accuracy": [90.0, 92.0],
+        "val_time": [1.0, 1.5], "test_loss": 1.1,
+        "test_accuracy": 64.0, "test_top5accuracy": 93.0, "test_time": 2.0,
+    }, 2)
+    assert dual["validation_selected_epoch"] == 2
+    assert dual["primary_test_accuracy_percent"] == 64.0
+    assert dual["test_evaluations"] == 1
+    assert dual["wall_s"] == 13.5
+    print("native final/validation-selected/test-selected semantics: exact")
+
+
 if __name__ == "__main__":
     check_fa_residual_transport()
     check_pepita_output_rule()
     check_ff_local_optimization()
     check_ep_energy_dynamics()
+    check_native_result_selection()
     print("ALL BASELINE SMOKE CHECKS PASSED")