Console Output

+ python tests/analyze_tuner_sample_history.py output/tuner/new_tuner_test -metric scaled_rmse -min_samples 2 --require_param_variation
Discovered 2 sample-history target(s).

=== Sample history 1/2: output/tuner/new_tuner_test/complex/cli_arm_bomex_20260730_200736_818125_3688748 ===
Tuner Sample History
  job: output/tuner/new_tuner_test/complex/cli_arm_bomex_20260730_200736_818125_3688748
  chunks: 1
  first_chunk: output/tuner/new_tuner_test/complex/cli_arm_bomex_20260730_200736_818125_3688748/sample_history/sample_history_000001.npz
  samples: 4
  params: 102
  cases: 2
  fields: 2
  observations: 10
  metrics: scaled_rmse, correlation, std_ratio, centered_rmse_norm, bias_norm
Checks performed
  - sample-history chunks exist and are sequentially numbered
  - every chunk contains required metadata, sample, and loss arrays
  - metadata arrays match exactly across chunks
  - schema version matches the current observation-axis format
  - sample IDs are unique and cover 0..sample_count-1
  - parameter and loss arrays have the expected shapes and finite values
  - observation metadata covers every case/field/window exactly once
  - at least 2 sample(s) are present
  - at least one tuned parameter varies across samples
  checks: OK
SVD (scaled_rmse)
  matrix_shape: 4 samples x 10 retained columns
  mode 1: singular=6.10319, explained=93.12%
    strongest observations:
      bomex/cloud_frac/w3:12600-15300s/scaled_rmse loading=-0.327
      bomex/cloud_frac/w1:7200-9900s/scaled_rmse loading=-0.327
      bomex/cloud_frac/w2:9900-12600s/scaled_rmse loading=-0.326
      arm/rcm/w1:68400-72000s/scaled_rmse loading=-0.326
      arm/cloud_frac/w1:68400-72000s/scaled_rmse loading=-0.324
    strongest parameter correlations:
      C8=+0.956
      C11=+0.270
  mode 2: singular=1.37261, explained=4.71%
    strongest observations:
      bomex/rcm/w3:12600-15300s/scaled_rmse loading=-0.744
      bomex/rcm/w1:7200-9900s/scaled_rmse loading=+0.458
      bomex/rcm/w2:9900-12600s/scaled_rmse loading=-0.343
      arm/cloud_frac/w1:68400-72000s/scaled_rmse loading=+0.187
      bomex/rcm/w4:15300-18000s/scaled_rmse loading=-0.155
    strongest parameter correlations:
      C11=-0.941
      C8=+0.236
  mode 3: singular=0.931153, explained=2.17%
    strongest observations:
      bomex/rcm/w1:7200-9900s/scaled_rmse loading=-0.746
      bomex/rcm/w4:15300-18000s/scaled_rmse loading=+0.418
      bomex/rcm/w3:12600-15300s/scaled_rmse loading=-0.331
      bomex/cloud_frac/w4:15300-18000s/scaled_rmse loading=+0.244
      bomex/rcm/w2:9900-12600s/scaled_rmse loading=-0.183
    strongest parameter correlations:
      C11=+0.206
      C8=-0.176

=== Sample history 2/2: output/tuner/new_tuner_test/simple/cli_bomex_20260730_200725_753191_3687632 ===
Tuner Sample History
  job: output/tuner/new_tuner_test/simple/cli_bomex_20260730_200725_753191_3687632
  chunks: 1
  first_chunk: output/tuner/new_tuner_test/simple/cli_bomex_20260730_200725_753191_3687632/sample_history/sample_history_000001.npz
  samples: 8
  params: 102
  cases: 1
  fields: 1
  observations: 1
  metrics: scaled_rmse, correlation, std_ratio, centered_rmse_norm, bias_norm
Checks performed
  - sample-history chunks exist and are sequentially numbered
  - every chunk contains required metadata, sample, and loss arrays
  - metadata arrays match exactly across chunks
  - schema version matches the current observation-axis format
  - sample IDs are unique and cover 0..sample_count-1
  - parameter and loss arrays have the expected shapes and finite values
  - observation metadata covers every case/field/window exactly once
  - at least 2 sample(s) are present
  - at least one tuned parameter varies across samples
  checks: OK
SVD (scaled_rmse)
  matrix_shape: 8 samples x 1 retained columns
  mode 1: singular=2.82843, explained=100.00%
    strongest observations:
      bomex/cloud_frac/w1:10800-21600s/scaled_rmse loading=+1.000
    strongest parameter correlations:
      C8=-0.999