Add more tests

tests/test_block_coordinate.py

@@ -3,7 +3,10 @@
 from pytest import _report_data_type, _test_prediction_method_with_different_types
 from scipy.sparse import csr_matrix
 
-from xcolumns.block_coordinate import predict_using_bc_with_0approx
+from xcolumns.block_coordinate import (
+    predict_optimizing_coverage_using_bc,
+    predict_using_bc_with_0approx,
+)
 from xcolumns.confusion_matrix import calculate_confusion_matrix
 from xcolumns.metrics import binary_recall_on_conf_matrix, macro_recall_on_conf_matrix
 from xcolumns.weighted_prediction import predict_optimizing_macro_recall, predict_top_k
@@ -91,3 +94,51 @@ def test_block_coordinate_with_different_types(generated_test_data):
     )
     assert bc_score >= top_k_score
     assert abs(opt_recall_score - bc_score) < 0.02
+
+
+def _run_block_coordinate_for_coverage(y_test, y_proba_test, k, init_y_pred):
+    _report_data_type(y_proba_test)
+    y_pred, meta = predict_optimizing_coverage_using_bc(
+        y_proba_test,
+        k,
+        return_meta=True,
+        seed=2024,
+        init_y_pred=init_y_pred,
+    )
+    print(f"  time={meta['time']}s")
+
+    assert type(y_pred) == type(y_proba_test)
+    assert y_pred.dtype == y_proba_test.dtype
+    if k > 0:
+        assert (y_pred.sum(axis=1) == k).all()
+    return (
+        calculate_confusion_matrix(y_test, y_pred, normalize=False, skip_tn=False),
+        y_pred,
+    )
+
+
+def test_block_coordinate_for_coverage(generated_test_data):
+    y_test = generated_test_data["y_test"]
+    y_proba_test = generated_test_data["y_proba_test"]
+    k = 3
+
+    # Run predict_top_k to get baseline classifier and initial prediction
+    top_k_y_pred = predict_top_k(y_proba_test, k)
+    top_k_C = calculate_confusion_matrix(
+        y_test, top_k_y_pred, normalize=False, skip_tn=False
+    )
+
+    conf_mats, y_preds = _test_prediction_method_with_different_types(
+        _run_block_coordinate_for_coverage,
+        (y_test, y_proba_test, k, top_k_y_pred),
+        test_torch=False,
+    )
+
+    # Compare with top-k
+    top_k_score = macro_recall_on_conf_matrix(*top_k_C)
+    bc_score = macro_recall_on_conf_matrix(*conf_mats[0])
+    opt_recall_score = macro_recall_on_conf_matrix(*opt_recall_C)
+    print(
+        f"Top-k score={top_k_score}, BC score={bc_score}, opt recall score={opt_recall_score}"
+    )
+    assert bc_score >= top_k_score

tests/test_metrics.py

@@ -2,6 +2,8 @@
 
 from xcolumns.confusion_matrix import calculate_confusion_matrix
 from xcolumns.metrics import (
+    binary_0_1_loss_on_conf_matrix,
+    binary_accuracy_on_conf_matrix,
     binary_balanced_accuracy_on_conf_matrix,
     binary_f1_score_on_conf_matrix,
     binary_fbeta_score_on_conf_matrix,
@@ -10,7 +12,27 @@
     binary_jaccard_score_on_conf_matrix,
     binary_precision_on_conf_matrix,
     binary_recall_on_conf_matrix,
+    coverage_on_conf_matrix,
+    hamming_loss_on_conf_matrix,
+    hamming_score_on_conf_matrix,
+    macro_balanced_accuracy_on_conf_matrix,
+    macro_f1_score_on_conf_matrix,
+    macro_fbeta_score_on_conf_matrix,
+    macro_gmean_on_conf_matrix,
+    macro_hmean_on_conf_matrix,
+    macro_jaccard_score_on_conf_matrix,
+    macro_precision_on_conf_matrix,
+    macro_recall_on_conf_matrix,
+    micro_balanced_accuracy_on_conf_matrix,
+    micro_f1_score_on_conf_matrix,
+    micro_fbeta_score_on_conf_matrix,
+    micro_gmean_on_conf_matrix,
+    micro_hmean_on_conf_matrix,
+    micro_jaccard_score_on_conf_matrix,
+    micro_precision_on_conf_matrix,
+    micro_recall_on_conf_matrix,
 )
+from xcolumns.types import Number
 from xcolumns.weighted_prediction import predict_top_k
 
 
@@ -26,6 +48,8 @@ def test_binary_metrics_on_conf_matrix(generated_test_data):
     )
 
     for metric in [
+        binary_0_1_loss_on_conf_matrix,
+        binary_accuracy_on_conf_matrix,
         binary_balanced_accuracy_on_conf_matrix,
         binary_fbeta_score_on_conf_matrix,
         binary_f1_score_on_conf_matrix,
@@ -41,3 +65,42 @@ def test_binary_metrics_on_conf_matrix(generated_test_data):
         assert isinstance(result, np.ndarray)
         assert result.shape == (y_proba_test.shape[1],)
         assert (0 <= result).all() and (result <= 1).all()
+
+
+def test_multilabel_metrics_on_conf_matrix(generated_test_data):
+    y_test = generated_test_data["y_test"]
+    y_proba_test = generated_test_data["y_proba_test"]
+    k = 3
+
+    # Run predict_top_k to get baseline classifier and initial prediction
+    top_k_y_pred = predict_top_k(y_proba_test, k)
+    top_k_C = calculate_confusion_matrix(
+        y_test, top_k_y_pred, normalize=False, skip_tn=False
+    )
+
+    for metric in [
+        micro_balanced_accuracy_on_conf_matrix,
+        micro_f1_score_on_conf_matrix,
+        micro_fbeta_score_on_conf_matrix,
+        micro_gmean_on_conf_matrix,
+        micro_hmean_on_conf_matrix,
+        micro_jaccard_score_on_conf_matrix,
+        micro_precision_on_conf_matrix,
+        micro_recall_on_conf_matrix,
+        macro_balanced_accuracy_on_conf_matrix,
+        macro_f1_score_on_conf_matrix,
+        macro_fbeta_score_on_conf_matrix,
+        macro_gmean_on_conf_matrix,
+        macro_hmean_on_conf_matrix,
+        macro_jaccard_score_on_conf_matrix,
+        macro_precision_on_conf_matrix,
+        macro_recall_on_conf_matrix,
+        coverage_on_conf_matrix,
+        hamming_loss_on_conf_matrix,
+        hamming_score_on_conf_matrix,
+    ]:
+        print(f"testing {metric.__name__}")
+        result = metric(*top_k_C)
+        print(f"  result={result}")
+        assert isinstance(result, Number)
+        assert result >= 0

tests/test_weighted_prediction.py

@@ -3,12 +3,14 @@
 from scipy.sparse import csr_matrix
 
 from xcolumns.confusion_matrix import calculate_confusion_matrix
+from xcolumns.metrics import jpv_inverse_propensities, label_priors
 from xcolumns.utils import *
 from xcolumns.weighted_prediction import (
     predict_log_weighted_per_instance,
     predict_optimizing_instance_precision,
     predict_optimizing_instance_propensity_scored_precision,
     predict_optimizing_macro_balanced_accuracy,
+    predict_optimizing_macro_recall,
     predict_power_law_weighted_per_instance,
     predict_weighted_per_instance,
 )
@@ -79,16 +81,19 @@ def test_prediction_optizming_macro_balanced_accuracy(generated_test_data):
     )
 
 
-# def test_wrapper_methods(generated_test_data):
-#     y_test = generated_test_data["y_test"]
-#     y_proba_test = generated_test_data["y_proba_test"]
+def test_wrapper_methods(generated_test_data):
+    y_test = generated_test_data["y_test"]
+    y_proba_test = generated_test_data["y_proba_test"]
 
-#     priors = y_test.mean(axis=0)
+    priors = label_priors(y_test)
+    inverse_propensities = jpv_inverse_propensities(y_test)
 
-#     for func in [
-#         predict_log_weighted_per_instance,
-#         predict_power_law_weighted_per_instance,
-#         predict_optimizing_instance_propensity_scored_precision,
-#         predict_optimizing_instance_precision,
-#     ]:
-#         func(y_proba_test, k=3)
+    y_pred = predict_optimizing_instance_precision(y_proba_test, k=3)
+    y_pred = predict_log_weighted_per_instance(y_proba_test, k=3, priors=priors)
+    y_pred = predict_power_law_weighted_per_instance(
+        y_proba_test, k=3, priors=priors, beta=0.5
+    )
+    y_pred = predict_optimizing_instance_propensity_scored_precision(
+        y_proba_test, k=3, inverse_propensities=inverse_propensities
+    )
+    y_pred = predict_optimizing_macro_recall(y_proba_test, k=3, priors=priors)