Merge ed66aa9 into 07a6faf

csrank/tests/test_choice_functions.py

@@ -10,7 +10,7 @@
 from csrank.experiments.constants import *
 from csrank.experiments.util import metrics_on_predictions
 from csrank.metrics_np import f1_measure, subset_01_loss, instance_informedness, auc_score
-from csrank.tests.test_ranking import check_leaner
+from csrank.tests.test_ranking import check_learner
 
 choice_metrics = {'F1Score': f1_measure, 'Informedness': instance_informedness, "AucScore": auc_score}
 optimizer = SGD(lr=1e-3, momentum=0.9, nesterov=True)
@@ -77,4 +77,4 @@ def test_choice_function_fixed(trivial_choice_problem, name):
               "batch_size": 32, "alpha": 0.5, "l1_ratio": 0.7, "tol": 1e-2, "C": 10, "n_mixtures": 10, "n_nests": 5,
               "regularization": "l2"}
     learner.set_tunable_parameters(**params)
-    check_leaner(learner, params, rtol, atol)
+    check_learner(learner, params, rtol, atol)

csrank/tests/test_discrete_choice.py

@@ -11,7 +11,7 @@
 from csrank.experiments.constants import *
 from csrank.experiments.util import metrics_on_predictions
 from csrank.metrics_np import categorical_accuracy_np, topk_categorical_accuracy_np, subset_01_loss
-from csrank.tests.test_ranking import check_leaner
+from csrank.tests.test_ranking import check_learner
 
 metrics = {'CategoricalAccuracy': categorical_accuracy_np, 'CategoricalTopK2': topk_categorical_accuracy_np(k=2)}
 optimizer = SGD(lr=1e-3, momentum=0.9, nesterov=True)
@@ -85,4 +85,4 @@ def test_discrete_choice_function_fixed(trivial_discrete_choice_problem, name):
               "batch_size": 32, "alpha": 0.5, "l1_ratio": 0.7, "tol": 1e-2, "C": 10, "n_mixtures": 10, "n_nests": 5,
               "regularization": "l2"}
     learner.set_tunable_parameters(**params)
-    check_leaner(learner, params, rtol, atol)
+    check_learner(learner, params, rtol, atol)

csrank/tests/test_fate.py

@@ -53,7 +53,7 @@ def fit(self, *args, **kwargs):
     assert grc.batch_size == params["batch_size"]
     rtol = 1e-2
     atol = 1e-4
-    assert np.isclose(grc.optimizer.get_config()['lr'], params["learning_rate"], rtol=rtol, atol=atol, equal_nan=False)
+    assert np.isclose(grc.optimizer.get_config()['learning_rate'], params["learning_rate"], rtol=rtol, atol=atol, equal_nan=False)
     config = grc.kernel_regularizer.get_config()
     val1 = np.isclose(config["l1"], params["reg_strength"], rtol=rtol, atol=atol, equal_nan=False)
     val2 = np.isclose(config["l2"], params["reg_strength"], rtol=rtol, atol=atol, equal_nan=False)

csrank/tests/test_ranking.py

@@ -35,7 +35,7 @@ def trivial_ranking_problem():
     return x, y_true
 
 
-def check_leaner(ranker, params, rtol=1e-2, atol=1e-4):
+def check_learner(ranker, params, rtol=1e-2, atol=1e-4):
     for key, value in params.items():
         if key in ranker.__dict__.keys():
             expected = ranker.__dict__[key]
@@ -47,7 +47,7 @@ def check_leaner(ranker, params, rtol=1e-2, atol=1e-4):
             else:
                 assert value == expected
         elif key == "learning_rate" and "optimizer" in ranker.__dict__.keys():
-            assert np.isclose(ranker.optimizer.get_config()['lr'], value, rtol=rtol, atol=atol, equal_nan=False)
+            assert np.isclose(ranker.optimizer.get_config()['learning_rate'], value, rtol=rtol, atol=atol, equal_nan=False)
         elif key == "reg_strength" and "kernel_regularizer" in ranker.__dict__.keys():
             config = ranker.kernel_regularizer.get_config()
             val1 = np.isclose(config["l1"], value, rtol=rtol, atol=atol, equal_nan=False)
@@ -86,4 +86,4 @@ def test_object_ranker_fixed(trivial_ranking_problem, ranker_name):
               "batch_size": 32, "alpha": 0.5, "l1_ratio": 0.7, "tol": 1e-2, "C": 10, "n_mixtures": 10, "n_nests": 5,
               "regularization": "l2"}
     ranker.set_tunable_parameters(**params)
-    check_leaner(ranker, params, rtol, atol)
+    check_learner(ranker, params, rtol, atol)

requirements.txt

@@ -9,8 +9,7 @@ psycopg2-binary>=2.7
 docopt>=0.6.0
 joblib>=0.9.4
 tqdm>=4.11.2
-# keras 2.3 renamed `lr` to `learning_rate`, we will deal with that in the tf2 transition
-keras>=2.1.5,<2.3
+keras>=2.3
 pymc3>=3.5
 theano>=1.0
 # Pick either CPU or GPU version of tensorflow: