A few bugs involved "tests.sh"

.gitignore

@@ -0,0 +1,6 @@
+.idea/
+.vscode/
+.ipynb_checkpoints
+*/.ipynb_checkpoints
+__pycache__
+*/__pycache__

agents/threshold_policies.py

@@ -201,6 +201,7 @@ def equality_of_opportunity_thresholds(group_predictions,
         sample_weight=group_weights[group])
 
     roc[group] = (fprs, np.nan_to_num(tprs), thresholds)
+    # roc[group] = (np.nan_to_num(fprs), np.nan_to_num(tprs), thresholds)
 
   def negative_reward(tpr_target):
     """Returns negative reward suitable for optimization by minimization."""

examples/college_admission_util_test.py

@@ -75,7 +75,8 @@ def test_example_configuration_runs(self):
     referred to in college_admission_config.gin configuration file.
     """
     gin.parse_config_file(
-        'third_party/py/fairness_gym/examples/config/'
+        # 'third_party/py/fairness_gym/examples/config/'
+        'examples/config/'
         'college_admission_config.gin')
     runner = runner_lib.Runner()
     runner.run()

examples/config/college_admission_config.gin

@@ -1,9 +1,17 @@
-import fairness_gym.agents.college_admission_jury
-import fairness_gym.environments.college_admission
-import fairness_gym.examples.college_admission_util
-import fairness_gym.metrics.error_metrics
-import fairness_gym.metrics.value_tracking_metrics
-import fairness_gym.runner_lib
+# import fairness_gym.agents.college_admission_jury
+# import fairness_gym.environments.college_admission
+# import fairness_gym.examples.college_admission_util
+# import fairness_gym.metrics.error_metrics
+# import fairness_gym.metrics.value_tracking_metrics
+# import fairness_gym.runner_lib
+
+import agents.college_admission_jury
+import environments.college_admission
+import examples.college_admission_util
+import metrics.error_metrics
+import metrics.value_tracking_metrics
+import runner_lib
+
 
 # Configure the runner.
 Runner.num_steps = 3000

examples/docs/college_admission_example.md

@@ -100,7 +100,7 @@ contexts, even if the agent is not designed to anticipate strategic manipulation
 To run the experiments for the college admissions scenario to recreate the plots
 from the KDD 2019 paper [4] run the following command:
 
-`python examples/college_admissions_main --num_steps=750 --burnin=10
+`python examples/college_admissions_main.py --num_steps=750 --burnin=10
 --epsilon_greedy=False --verbose=True --noisy_dist='gaussian'
 --noisy_features=False --noisy_threshold=False --feature_mu='0.5','0.5'
 --output_dir="path/to/output_dir/kdd"`

examples/docs/lending_example.md

@@ -56,7 +56,7 @@ over the course of the simulation. Precision and Recall stratified by group.
 
 ## Interpreting the plots
 
-Running examples/lending_experiment_main.py reproduces the plots for the KDD
+Running examples/lending_experiments_main.py reproduces the plots for the KDD
 workshop paper [3].
 
 The cumulative loans given to each group are plotted over the course of the

requirements.txt

@@ -12,3 +12,7 @@ simplejson
 sklearn
 statsmodels
 tqdm
+
+gin-config==0.1.1
+mock
+scikit-learn==0.20.2  # better not to use 0.19.1