!116 fixed config verify issue.

Merge pull request !116 from liuluobin/master

mindarmour/privacy/evaluation/_check_config.py

@@ -15,6 +15,8 @@
 Verify attack config
 """
 
+import numpy as np
+
 from mindarmour.utils._check_param import check_param_type
 from mindarmour.utils.logger import LogUtil
 
@@ -138,7 +140,6 @@ _VALID_CONFIG_CHECKLIST = {
         "min_impurity_decrease": [_is_non_negative_float],
         "min_impurity_split": [{None}, _is_positive_float],
         "bootstrap": [{True, False}],
-        "oob_scroe": [{True, False}],
         "n_jobs": [_is_positive_int, {None}],
         "random_state": None,
         "verbose": [_is_non_negative_int],
@@ -182,7 +183,7 @@ def _check_config(attack_config, config_checklist):
         for param_key in params.keys():
             param_value = params[param_key]
             candidate_values = config_checklist[method][param_key]
-            check_param_type('param_value', param_value, list)
+            check_param_type('param_value', param_value, (list, tuple, np.ndarray))
 
             if candidate_values is None:
                 continue

mindarmour/privacy/evaluation/membership_inference.py

@@ -39,7 +39,7 @@ def _eval_info(pred, truth, option):
     Args:
         pred (numpy.ndarray): Predictions for each sample.
         truth (numpy.ndarray): Ground truth for each sample.
-        option(str): Type of evaluation indicators; Possible
+        option (str): Type of evaluation indicators; Possible
             values are 'precision', 'accuracy' and 'recall'.
 
     Returns:
@@ -77,7 +77,7 @@ def _softmax_cross_entropy(logits, labels):
 
     Args:
         logits (numpy.ndarray): Numpy array of shape(N, C).
-        labels (numpy.ndarray): Numpy array of shape(N, )
+        labels (numpy.ndarray): Numpy array of shape(N, ).
 
     Returns:
         numpy.ndarray: Numpy array of shape(N, ), containing loss value for each vector in logits.
@@ -136,7 +136,7 @@ class MembershipInference:
 
     def train(self, dataset_train, dataset_test, attack_config):
         """
-        Depending on the configuration, use the input data set to train the attack model.
+        Depending on the configuration, use the input dataset to train the attack model.
         Save the attack model to self._attack_list.
 
         Args:
@@ -148,10 +148,13 @@ class MembershipInference:
                 The support methods are knn, lr, mlp and rf, and the params of each method
                 must within the range of changeable parameters. Tips of params implement
                 can be found in
-                "https://scikit-learn.org/0.16/modules/generated/sklearn.grid_search.GridSearchCV.html".
+                https://scikit-learn.org/stable/modules/generated/sklearn.neighbors.KNeighborsClassifier.html
+                https://scikit-learn.org/stable/modules/generated/sklearn.linear_model.LogisticRegression.html
+                https://scikit-learn.org/stable/modules/generated/sklearn.ensemble.RandomForestClassifier.html
+                https://scikit-learn.org/stable/modules/generated/sklearn.neural_network.MLPRegressor.html
 
         Raises:
-            KeyError: If any config in attack_config doesn't have keys {"method", "params"}
+            KeyError: If any config in attack_config doesn't have keys {"method", "params"}.
             NameError: If the method(case insensitive) in attack_config is not in ["lr", "knn", "rf", "mlp"].
         """
         check_param_type("dataset_train", dataset_train, Dataset)