relocate python/paddle/fluid/regularizer.py (#53106)

* relocate regularizer.py

* fix bug

* fix bug

* fix bug

* relocate the import

* replace _regularization_coeff with coeff

* remove the L1DecayRegularizer and L2DecayRegularizer

python/paddle/distributed/fleet/meta_optimizers/dgc_optimizer.py

@@ -24,6 +24,7 @@
 from paddle.fluid.optimizer import Momentum, Optimizer
 from paddle.framework import core, in_dygraph_mode
 from paddle.nn.clip import ClipGradByNorm, append_gradient_clip_ops
+from paddle.regularizer import L1Decay, L2Decay
 from paddle.static import create_global_var
 
 
@@ -99,8 +100,7 @@ def _get_regularization_param(self, regularization):
         regular_coeff = 0.0
 
         if regularization is not None:
-            regular_coeff = regularization._regularization_coeff
-            from paddle.fluid.regularizer import L1Decay, L2Decay
+            regular_coeff = regularization._coeff
 
             if isinstance(regularization, L1Decay):
                 regular_type = 1

python/paddle/fluid/__init__.py

@@ -58,7 +58,6 @@
 from . import optimizer
 from . import backward
 from .backward import gradients
-from . import regularizer
 from . import incubate
 from .param_attr import ParamAttr, WeightNormParamAttr
 from .data_feeder import DataFeeder
@@ -116,7 +115,6 @@
         'nets',
         'optimizer',
         'backward',
-        'regularizer',
         'LoDTensor',
         'LoDTensorArray',
         'CPUPlace',

python/paddle/fluid/contrib/optimizer.py

@@ -12,8 +12,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 from paddle.fluid.optimizer import Optimizer
-from paddle.fluid.regularizer import L1DecayRegularizer
-from paddle.fluid.regularizer import L2DecayRegularizer
+from paddle.regularizer import L1Decay
+from paddle.regularizer import L2Decay
 from paddle.fluid import core
 from paddle.fluid import framework
 from paddle.fluid.framework import program_guard
@@ -117,7 +117,7 @@ def __init__(
     ):
         assert learning_rate is not None
         assert momentum is not None
-        predicate = lambda regular: isinstance(regular, L2DecayRegularizer)
+        predicate = lambda regular: isinstance(regular, L2Decay)
         py_regular = None if predicate(regularization) else regularization
         super().__init__(
             learning_rate=learning_rate,
@@ -131,9 +131,9 @@ def __init__(
         self._use_nesterov = bool(use_nesterov)
         self._regularization_method = ""
         self._regularization_coeff = 0
-        if isinstance(regularization, L2DecayRegularizer):
+        if isinstance(regularization, L2Decay):
             self._regularization_method = "l2_decay"
-            self._regularization_coeff = regularization._regularization_coeff
+            self._regularization_coeff = regularization._coeff
         self._multi_precision = multi_precision
         self._rescale_grad = rescale_grad
         self._master_weights = {}

python/paddle/fluid/layers/learning_rate_scheduler.py

@@ -413,7 +413,7 @@ def piecewise_decay(boundaries, values):
               optimizer = fluid.optimizer.Momentum(
                   momentum=0.9,
                   learning_rate=fluid.layers.piecewise_decay(boundaries=boundaries, values=values),
-                  regularization=fluid.regularizer.L2Decay(1e-4))
+                  regularization=paddle.regularizer.L2Decay(1e-4))
 
 
     """

python/paddle/fluid/param_attr.py

@@ -13,7 +13,7 @@
 # limitations under the License.
 
 import paddle
-from .regularizer import WeightDecayRegularizer
+from paddle.regularizer import WeightDecayRegularizer
 from paddle.fluid.data_feeder import check_type
 
 __all__ = [