check for valid quantiles for QPD, do not remove price ratio feature …

…in integration tests

cyclic_boosting/quantile_matching.py

@@ -46,6 +46,9 @@ def __init__(
         else:
             raise Exception("Invalid version.")
 
+        if (qv_low > qv_median) or (qv_high < qv_median):
+            raise ValueError("The SPT values need to be monotonically increasing.")
+
         self.l = l
 
         self.c = self.phi.ppf(1 - alpha)

tests/test_integration.py

@@ -254,8 +254,8 @@ def test_nbinom_regression_ndarray(prepare_data, default_features, feature_prope
 
 @pytest.fixture(scope="function")
 def cb_exponential_regressor_model(features, feature_properties):
-    features = features
-    features.remove("price_ratio")
+    features_noprice = features.copy()
+    features_noprice.remove("price_ratio")
     price_features = [
         "L_ID",
         "PG_ID_1",
@@ -278,7 +278,7 @@ def cb_exponential_regressor_model(features, feature_properties):
 
     CB_pipeline = pipeline_CBExponential(
         feature_properties=feature_properties,
-        standard_feature_groups=features,
+        standard_feature_groups=features_noprice,
         external_feature_groups=price_features,
         external_colname="price_ratio",
         observers=plobs,
@@ -464,7 +464,7 @@ def test_multiplicative_quantile_regression_median(is_plot, prepare_data, featur
     np.testing.assert_almost_equal(quantile_acc, 0.5043, 3)
 
     mad = np.nanmean(np.abs(y - yhat))
-    np.testing.assert_almost_equal(mad, 1.6575, 3)
+    np.testing.assert_almost_equal(mad, 1.6559, 3)
 
 
 def test_multiplicative_quantile_regression_90(is_plot, prepare_data, features, feature_properties):
@@ -506,13 +506,16 @@ def test_multiplicative_quantile_regression_pdf_J_QPD_S(is_plot, prepare_data, f
     cdf_truth_list = []
     n_samples = len(X)
     for i in range(n_samples):
-        j_qpd_s = J_QPD_S(0.2, quantile_values[0, i], quantile_values[1, i], quantile_values[2, i])
+        try:
+            j_qpd_s = J_QPD_S(0.2, quantile_values[0, i], quantile_values[1, i], quantile_values[2, i])
+        except ValueError:
+            continue
         np.testing.assert_almost_equal(j_qpd_s.ppf(0.2), quantile_values[0, i], 3)
         np.testing.assert_almost_equal(j_qpd_s.ppf(0.5), quantile_values[1, i], 3)
         np.testing.assert_almost_equal(j_qpd_s.ppf(0.8), quantile_values[2, i], 3)
         if i == 24:
-            np.testing.assert_almost_equal(j_qpd_s.ppf(0.1), 0.592, 3)
-            np.testing.assert_almost_equal(j_qpd_s.ppf(0.9), 5.783, 3)
+            np.testing.assert_almost_equal(j_qpd_s.ppf(0.1), 0.457, 3)
+            np.testing.assert_almost_equal(j_qpd_s.ppf(0.9), 5.509, 3)
 
             if is_plot:
                 plt.plot([0.2, 0.5, 0.8], [quantile_values[0, i], quantile_values[1, i], quantile_values[2, i]], "ro")
@@ -552,9 +555,9 @@ def test_multiplicative_quantile_regression_spline(is_plot, prepare_data, featur
 
     i = 24
     spl_fit = quantile_fit_spline(quantiles, quantile_values[:, i])
-    np.testing.assert_almost_equal(spl_fit(0.2), 0.679, 3)
-    np.testing.assert_almost_equal(spl_fit(0.5), 2.202, 3)
-    np.testing.assert_almost_equal(spl_fit(0.8), 4.297, 3)
+    np.testing.assert_almost_equal(spl_fit(0.2), 0.527, 3)
+    np.testing.assert_almost_equal(spl_fit(0.5), 2.193, 3)
+    np.testing.assert_almost_equal(spl_fit(0.8), 4.21, 3)
 
     if is_plot:
         plt.plot(quantiles, quantile_values[:, i], "ro")
@@ -587,9 +590,9 @@ def test_multiplicative_quantile_regression_pdf_gamma(is_plot, prepare_data, fea
     for i in range(n_samples):
         if i == 24:
             gamma_fit = quantile_fit_gamma(quantiles, quantile_values[:, i])
-            np.testing.assert_almost_equal(gamma_fit(0.2), 0.829, 3)
-            np.testing.assert_almost_equal(gamma_fit(0.5), 2.049, 3)
-            np.testing.assert_almost_equal(gamma_fit(0.8), 4.147, 3)
+            np.testing.assert_almost_equal(gamma_fit(0.2), 0.779, 3)
+            np.testing.assert_almost_equal(gamma_fit(0.5), 1.986, 3)
+            np.testing.assert_almost_equal(gamma_fit(0.8), 4.097, 3)
 
             if is_plot:
                 plt.plot(quantiles, quantile_values[:, i], "ro")
@@ -824,4 +827,4 @@ def test_classification_logloss(is_plot, prepare_data, cb_classifier_logloss_mod
     yhat = CB_est.predict(X.copy())
 
     mad = np.nanmean(np.abs(y - yhat))
-    np.testing.assert_almost_equal(mad, 0.408, 3)
+    np.testing.assert_almost_equal(mad, 0.404, 3)