Signed-off-by: weijingchen <talkingwallace@sohu.com>

Fix examples & fix sbt label check

Signed-off-by: cwj <talkingwallace@sohu.com>

python/fate/ml/ensemble/algo/secureboost/hetero/guest.py

@@ -196,15 +196,19 @@ def _init_sample_scores(self, ctx: Context, label, train_data: DataFrame):
                 self._accumulate_scores = self._loss_func.initialize(label)
 
     def _check_label(self, label: DataFrame):
-        label_df = label.as_pd_df()[label.schema.label_name]
+
+        train_data_binarized_label = label.get_dummies()
+        labels = [int(label_name.split("_")[1]) for label_name in train_data_binarized_label.columns]
+        label_set = set(labels)
+
+
         if self.objective == MULTI_CE:
 
             if self.num_class is None or self.num_class <= 2:
                 raise ValueError(
                     f"num_class should be set and greater than 2 for multi:ce objective, but got {self.num_class}"
                 )
 
-            label_set = set(np.unique(label_df))
             if len(label_set) > self.num_class:
                 raise ValueError(
                     f"num_class should be greater than or equal to the number of unique label in provided train data, but got {self.num_class} and {len(label_set)}"
@@ -215,7 +219,6 @@ def _check_label(self, label: DataFrame):
                 )
 
         elif self.objective == BINARY_BCE:
-            label_set = set(np.unique(label_df))
             assert len(label_set) == 2, f"binary classification task should have 2 unique label, but got {label_set}"
             assert (
                 0 in label_set and 1 in label_set

python/fate/ml/ensemble/algo/secureboost/test/test_hetero_sbt_multi.py

@@ -33,7 +33,7 @@ def create_ctx(local):
     console_handler.setFormatter(formatter)
 
     logger.addHandler(console_handler)
-    computing = CSession()
+    computing = CSession(data_dir='./')
     return Context(
         computing=computing, federation=StandaloneFederation(computing, name, local, [guest, host, arbiter])
     )

python/fate/ml/nn/test/test_hetero_nn_sshe.py

@@ -17,8 +17,8 @@ def get_current_datetime_str():
 
 def create_ctx(local, context_name):
     from fate.arch import Context
-    from fate.arch.computing.standalone import CSession
-    from fate.arch.federation.standalone import StandaloneFederation
+    from fate.arch.computing.backends.standalone import CSession
+    from fate.arch.federation.backends.standalone import StandaloneFederation
     import logging
 
     # prepare log
@@ -49,20 +49,27 @@ def set_seed(seed):
             t.backends.cudnn.deterministic = True
             t.backends.cudnn.benchmark = False
 
-    set_seed(42)
-
     batch_size = 64
     epoch = 10
-    guest_bottom = t.nn.Linear(10, 4)
+    guest_bottom = t.nn.Linear(10, 10)
     guest_top = t.nn.Sequential(
-                                 t.nn.Linear(4, 1),
+                                 t.nn.Linear(10, 1),
                                  t.nn.Sigmoid()
                                )
-    host_bottom = t.nn.Linear(20, 4)
+    host_bottom = t.nn.Linear(20, 10)
 
     # # make random fake data
     sample_num = 569
 
+    args = TrainingArguments(
+        num_train_epochs=1,
+        per_device_train_batch_size=256,
+        logging_strategy='epoch',
+        no_cuda=True,
+        log_level='debug',
+        disable_tqdm=False
+    )
+
     if party == "guest":
 
         from fate.ml.evaluation.metric_base import MetricEnsemble
@@ -80,21 +87,13 @@ def set_seed(seed):
             top_model=guest_top,
             bottom_model=guest_bottom,
             agglayer_arg=SSHEArgument(
-                guest_in_features=4,
-                host_in_features=4,
-                out_features=4,
-                layer_lr=0.01
+                guest_in_features=10,
+                host_in_features=10,
+                out_features=10
             )
         )
-        model
         optimizer = t.optim.Adam(model.parameters(), lr=0.01)
 
-        args = TrainingArguments(
-            num_train_epochs=5,
-            per_device_train_batch_size=16,
-            no_cuda=True,
-            disable_tqdm=False
-        )
         trainer = HeteroNNTrainerGuest(
             ctx=ctx,
             model=model,
@@ -106,10 +105,15 @@ def set_seed(seed):
             compute_metrics=MetricEnsemble().add_metric(MultiAccuracy())
         )
         trainer.train()
-        pred = trainer.predict(dataset)
+        pred_0 = trainer.predict(dataset)
         # # compute auc
         from sklearn.metrics import roc_auc_score
-        print(roc_auc_score(pred.label_ids, pred.predictions))
+        print(roc_auc_score(pred_0.label_ids, pred_0.predictions))
+
+        pred_1 = trainer.predict(dataset)
+        # # compute auc
+        from sklearn.metrics import roc_auc_score
+        print(roc_auc_score(pred_1.label_ids, pred_1.predictions))
 
     elif party == "host":
 
@@ -123,21 +127,13 @@ def set_seed(seed):
         model = HeteroNNModelHost(
             bottom_model=host_bottom,
             agglayer_arg=SSHEArgument(
-                guest_in_features=4,
-                host_in_features=4,
-                out_features=4,
-                layer_lr=0.01
+                guest_in_features=10,
+                host_in_features=10,
+                out_features=10
             )
         )
         optimizer = t.optim.Adam(model.parameters(), lr=0.01)
 
-        args = TrainingArguments(
-            num_train_epochs=5,
-            per_device_train_batch_size=16,
-            no_cuda=True,
-            disable_tqdm=False
-        )
-
         trainer = HeteroNNTrainerHost(
             ctx=ctx,
             model=model,
@@ -148,4 +144,4 @@ def set_seed(seed):
         )
         trainer.train()
         trainer.predict(dataset)
-
+        trainer.predict(dataset)