Rename dnn_units to 'hidden_units'

DataCanvasIO · Sep 9, 2020 · d1201d9 · d1201d9
1 parent 07ca646
commit d1201d9
Show file tree

Hide file tree

Showing 9 changed files with 49 additions and 43 deletions.
diff --git a/deeptables/models/config.py b/deeptables/models/config.py
@@ -82,8 +82,8 @@ def __new__(cls,
                 optimizer='auto',
                 loss='auto',
                 dnn_params={
-                    'dnn_units': ((128, 0, False), (64, 0, False)),
-                    'dnn_activation': 'relu',
+                    'hidden_units': ((128, 0, False), (64, 0, False)),
+                    'activation': 'relu',
                 },
                 autoint_params={
                     'num_attention': 3,

diff --git a/deeptables/models/deeptable.py b/deeptables/models/deeptable.py
@@ -159,7 +159,7 @@ class DeepTable:
 
             loss: str or object, (default='auto')
 
-            dnn_params: dict, (default={'dnn_units': ((128, 0, False), (64, 0, False)),
+            dnn_params: dict, (default={'hidden_units': ((128, 0, False), (64, 0, False)),
                                         'dnn_activation': 'relu'})
 
             autoint_params:dict, (default={'num_attention': 3,'num_heads': 1,

diff --git a/deeptables/models/hyper_dt.py b/deeptables/models/hyper_dt.py
@@ -29,14 +29,13 @@ def _on_params_ready(self):
 
 
 class DTFit(ModuleSpace):
-    def __init__(self, batch_size=128, epochs=1000, space=None, name=None, **hyperparams):
+    def __init__(self, batch_size=128, epochs=None, space=None, name=None, **hyperparams):
         if batch_size is None:
             batch_size = Choice([128, 256, 512])
         hyperparams['batch_size'] = batch_size
 
-        if epochs is None:
-            epochs = 1000
-        hyperparams['epochs'] = epochs
+        if epochs is not None:
+            hyperparams['epochs'] = epochs
 
         ModuleSpace.__init__(self, space, name, **hyperparams)
         self.space.fit_params = self
@@ -52,11 +51,11 @@ def _on_params_ready(self):
 
 
 class DnnModule(ModuleSpace):
-    def __init__(self, dnn_units=None, reduce_factor=None, dnn_dropout=None, use_bn=None, dnn_layers=None,
+    def __init__(self, hidden_units=None, reduce_factor=None, dnn_dropout=None, use_bn=None, dnn_layers=None,
                  activation=None, space=None, name=None, **hyperparams):
-        if dnn_units is None:
-            dnn_units = Choice([100, 200, 300, 500, 800, 1000])
-        hyperparams['dnn_units'] = dnn_units
+        if hidden_units is None:
+            hidden_units = Choice([100, 200, 300, 500, 800, 1000])
+        hyperparams['hidden_units'] = hidden_units
 
         if reduce_factor is None:
             reduce_factor = Choice([1, 0.8, 0.5])
@@ -82,15 +81,15 @@ def __init__(self, dnn_units=None, reduce_factor=None, dnn_dropout=None, use_bn=
 
     def _compile(self):
         dnn_layers = self.param_values['dnn_layers']
-        dnn_units = []
+        hidden_units = []
         for i in range(0, dnn_layers):
-            dnn_units.append(
-                (int(self.param_values['dnn_units'] * 1 if i == 0 else (
-                        self.param_values['dnn_units'] * (self.param_values['reduce_factor'] ** i))),
+            hidden_units.append(
+                (int(self.param_values['hidden_units'] * 1 if i == 0 else (
+                        self.param_values['hidden_units'] * (self.param_values['reduce_factor'] ** i))),
                  self.param_values['dnn_dropout'],
                  self.param_values['use_bn']))
         dnn_params = {
-            'dnn_units': dnn_units,
+            'hidden_units': hidden_units,
             'dnn_activation': self.param_values['activation'],
         }
         self.space.DT_Module.config = self.space.DT_Module.config._replace(dnn_params=dnn_params)
@@ -225,7 +224,7 @@ def mini_dt_space():
             apply_class_weight=Bool(),
             earlystopping_patience=Choice([1, 3, 5])
         )
-        dnn = DnnModule(dnn_units=Choice([100, 200]),
+        dnn = DnnModule(hidden_units=Choice([100, 200]),
                         reduce_factor=Choice([1, 0.8]),
                         dnn_dropout=Choice([0, 0.3]),
                         use_bn=Bool(),
@@ -261,7 +260,7 @@ def mini_dt_space():
     # optimizer='auto',
     # loss='auto',
     # dnn_params={
-    #     'dnn_units': ((128, 0, False), (64, 0, False)),
+    #     'hidden_units': ((128, 0, False), (64, 0, False)),
     #     'dnn_activation': 'relu',
     # },
     # autoint_params={

diff --git a/docs/source/examples.md b/docs/source/examples.md
@@ -66,7 +66,7 @@ conf = ModelConfig(
     metrics=['RootMeanSquaredError'], 
     nets=['dnn_nets'],
     dnn_params={
-        'dnn_units': ((256, 0.3, True), (256, 0.3, True)),
+        'hidden_units': ((256, 0.3, True), (256, 0.3, True)),
         'dnn_activation': 'relu',
     },
     earlystopping_patience=5,

diff --git a/docs/source/model_config.md b/docs/source/model_config.md
@@ -215,6 +215,12 @@ str, (default=`'add'`)
 ### output_use_bias
 bool, (default=`True`)
 
+### apply_class_weight
+bool, (default='False')
+
+Whether to calculate the weight of each class automatically. This can be useful to tell the model to "pay more attention" to samples from an under-represented class.
+
+
 ### optimizer
 str(name of optimizer) or optimizer instance or 'auto', (default=`'auto'`)
 
@@ -259,7 +265,7 @@ dictionary
 Only usable when 'dnn_nets' or a component using 'dnn' like 'pnn_nets','dcn_nets' included in [nets].
 ```
 {
-    'dnn_units': ((128, 0, False), (64, 0, False)),
+    'hidden_units': ((128, 0, False), (64, 0, False)),
     'dnn_activation': 'relu'}
 )
 ```

diff --git a/examples/dt_regression.ipynb b/examples/dt_regression.ipynb
@@ -98,7 +98,7 @@
     "    metrics=['RootMeanSquaredError'], \n",
     "    nets=['dnn_nets'],\n",
     "    dnn_params={\n",
-    "        'dnn_units': ((256, 0.3, True), (256, 0.3, True)),\n",
+    "        'hidden_units': ((256, 0.3, True), (256, 0.3, True)),\n",
     "        'dnn_activation': 'relu',\n",
     "    },\n",
     "    earlystopping_patience=5,\n",

diff --git a/tests/models/hyper_dt_test.py b/tests/models/hyper_dt_test.py
@@ -23,7 +23,7 @@ def bankdata(self):
                       reward_metric='accuracy',
                       max_trails=3,
                       dnn_params={
-                          'dnn_units': ((256, 0, False), (256, 0, False)),
+                          'hidden_units': ((256, 0, False), (256, 0, False)),
                           'dnn_activation': 'relu',
                       },
                       )
@@ -49,13 +49,14 @@ def test_default_dt_space(self):
         space = default_dt_space()
         space.random_sample()
         assert space.Module_DnnModule_1.param_values['dnn_layers'] == len(
-            space.DT_Module.config.dnn_params['dnn_units'])
-        assert space.Module_DnnModule_1.param_values['dnn_units'] == space.DT_Module.config.dnn_params['dnn_units'][0][
-            0]
+            space.DT_Module.config.dnn_params['hidden_units'])
+        assert space.Module_DnnModule_1.param_values['hidden_units'] == \
+               space.DT_Module.config.dnn_params['hidden_units'][0][
+                   0]
         assert space.Module_DnnModule_1.param_values['dnn_dropout'] == \
-               space.DT_Module.config.dnn_params['dnn_units'][0][
+               space.DT_Module.config.dnn_params['hidden_units'][0][
                    1]
-        assert space.Module_DnnModule_1.param_values['use_bn'] == space.DT_Module.config.dnn_params['dnn_units'][0][
+        assert space.Module_DnnModule_1.param_values['use_bn'] == space.DT_Module.config.dnn_params['hidden_units'][0][
             2]
 
     def test_hyper_dt(self):
@@ -64,7 +65,7 @@ def test_hyper_dt(self):
                       callbacks=[SummaryCallback()],
                       reward_metric='accuracy',
                       dnn_params={
-                          'dnn_units': ((256, 0, False), (256, 0, False)),
+                          'hidden_units': ((256, 0, False), (256, 0, False)),
                           'dnn_activation': 'relu',
                       },
                       cache_preprocessed_data=True,
@@ -77,11 +78,11 @@ def test_hyper_dt(self):
 
         y = np.random.randint(0, 2, size=(100), dtype='int')
         df = pd.DataFrame({'x1': x1, 'x2': x2, 'x3': x3, 'x4': x4})
-        hdt.search(df, y, df, y, max_trails=3, )
+        hdt.search(df, y, df, y, max_trails=3, epochs=1)
         assert hdt.best_model
         best_trial = hdt.get_best_trail()
 
-        estimator = hdt.final_train(best_trial.space_sample, df, y, cross_validation=True, num_folds=3)
+        estimator = hdt.final_train(best_trial.space_sample, df, y, epochs=1)
         score = estimator.predict(df)
         result = estimator.evaluate(df, y)
         assert len(score) == 100

diff --git a/tests/utils/batch_trainer_test.py b/tests/utils/batch_trainer_test.py
@@ -16,7 +16,7 @@ class Test_Batch_Trainer:
     def test_run_binary_heart_disease_CV(self):
         data = dsutils.load_heart_disease_uci()
         conf = deeptable.ModelConfig(
-            dnn_params={'dnn_units': ((256, 0, False), (256, 0, False)),
+            dnn_params={'hidden_units': ((256, 0, False), (256, 0, False)),
                         'dnn_activation': 'relu'},
             fixed_embedding_dim=False,
             embeddings_output_dim=0,
@@ -46,7 +46,7 @@ def test_run_binary_heart_disease_CV(self):
     def test_run_lgbm(self):
         data = dsutils.load_adult().head(1000)
         conf = deeptable.ModelConfig(
-            dnn_params={'dnn_units': ((256, 0, False), (256, 0, False)),
+            dnn_params={'hidden_units': ((256, 0, False), (256, 0, False)),
                         'dnn_activation': 'relu'},
             fixed_embedding_dim=False,
             embeddings_output_dim=0,
@@ -72,7 +72,7 @@ def test_run_lgbm(self):
     def test_run_catboost(self):
         data = dsutils.load_adult().head(1000)
         conf = deeptable.ModelConfig(
-            dnn_params={'dnn_units': ((256, 0, False), (256, 0, False)),
+            dnn_params={'hidden_units': ((256, 0, False), (256, 0, False)),
                         'dnn_activation': 'relu'},
             fixed_embedding_dim=False,
             embeddings_output_dim=0,
@@ -97,7 +97,7 @@ def test_run_catboost(self):
     def test_run_binary(self):
         data = dsutils.load_adult().head(1000)
         conf = deeptable.ModelConfig(
-            dnn_params={'dnn_units': ((256, 0, False), (256, 0, False)),
+            dnn_params={'hidden_units': ((256, 0, False), (256, 0, False)),
                         'dnn_activation': 'relu'},
             fixed_embedding_dim=False,
             embeddings_output_dim=0,
@@ -124,7 +124,7 @@ def test_run_regression(self):
         df_train.insert(df_train.shape[1], 'target', boston_dataset.target)
 
         conf = deeptable.ModelConfig(
-            dnn_params={'dnn_units': ((256, 0, False), (256, 0, False)),
+            dnn_params={'hidden_units': ((256, 0, False), (256, 0, False)),
                         'dnn_activation': 'relu'},
             fixed_embedding_dim=False,
             embeddings_output_dim=0,
@@ -154,7 +154,7 @@ def test_run_regression(self):
     def test_run_multiclass(self):
         data = dsutils.load_glass_uci()
         conf = deeptable.ModelConfig(
-            # dnn_units=((256, 0, False), (128, 0, False)),
+            # hidden_units=((256, 0, False), (128, 0, False)),
             # dnn_activation='relu',
             fixed_embedding_dim=False,
             embeddings_output_dim=0,
@@ -180,7 +180,7 @@ def test_run_multiclass(self):
     def test_run_cross_validation(self):
         data = dsutils.load_adult().head(1000)
         conf = deeptable.ModelConfig(
-            # dnn_units=((256, 0, False), (128, 0, False)),
+            # hidden_units=((256, 0, False), (128, 0, False)),
             # dnn_activation='relu',
             fixed_embedding_dim=False,
             embeddings_output_dim=0,
@@ -241,7 +241,7 @@ def test_get_models(self):
     def test_get_models_retian_single_model(self):
         data = dsutils.load_adult().head(1000)
         conf = deeptable.ModelConfig(
-            # dnn_units=((256, 0, False), (128, 0, False)),
+            # hidden_units=((256, 0, False), (128, 0, False)),
             # dnn_activation='relu',
             fixed_embedding_dim=False,
             embeddings_output_dim=0,
@@ -278,7 +278,7 @@ def test_get_models_retian_single_model(self):
     def test_ensemble_predict_proba(self):
         data = dsutils.load_adult().head(1000)
         conf = deeptable.ModelConfig(
-            # dnn_units=((256, 0, False), (128, 0, False)),
+            # hidden_units=((256, 0, False), (128, 0, False)),
             # dnn_activation='relu',
             fixed_embedding_dim=False,
             embeddings_output_dim=0,
@@ -305,7 +305,7 @@ def test_ensemble_predict_proba(self):
     def test_probe_evaluation(self):
         data = dsutils.load_adult().head(1000)
         conf = deeptable.ModelConfig(
-            # dnn_units=((256, 0, False), (128, 0, False)),
+            # hidden_units=((256, 0, False), (128, 0, False)),
             # dnn_activation='relu',
             fixed_embedding_dim=False,
             embeddings_output_dim=0,
@@ -329,7 +329,7 @@ def test_probe_evaluation(self):
     def test_zero_testset(self):
         data = dsutils.load_adult().head(1000)
         conf = deeptable.ModelConfig(
-            # dnn_units=((256, 0, False), (128, 0, False)),
+            # hidden_units=((256, 0, False), (128, 0, False)),
             # dnn_activation='relu',
             fixed_embedding_dim=False,
             embeddings_output_dim=0,
@@ -354,7 +354,7 @@ def test_zero_testset(self):
     def test_zero_testset_cross_validation(self):
         data = dsutils.load_adult().head(1000)
         conf = deeptable.ModelConfig(
-            # dnn_units=((256, 0, False), (128, 0, False)),
+            # hidden_units=((256, 0, False), (128, 0, False)),
             # dnn_activation='relu',
             fixed_embedding_dim=False,
             embeddings_output_dim=0,
@@ -412,7 +412,7 @@ def test_multi_config(self):
     def test_leaderboard(self):
         data = dsutils.load_adult().head(1000)
         conf = deeptable.ModelConfig(
-            # dnn_units=((256, 0, False), (128, 0, False)),
+            # hidden_units=((256, 0, False), (128, 0, False)),
             # dnn_activation='relu',
             fixed_embedding_dim=False,
             embeddings_output_dim=0,

diff --git a/tests/models/shap_test.py → tests/utils/shap_test.py b/tests/models/shap_test.py → tests/utils/shap_test.py