rm const decay

configs/vqa/re/layoutxlm.yml

@@ -36,7 +36,7 @@ Optimizer:
   lr:
     learning_rate: 0.00005
   regularizer:
-    name: Const
+    name: L2
     factor: 0.00000
     
 PostProcess:

configs/vqa/ser/layoutlm.yml

@@ -38,7 +38,7 @@ Optimizer:
     epochs: *epoch_num
     warmup_epoch: 2
   regularizer:
-    name: Const
+    name: L2
     factor: 0.00000
     
 PostProcess:

configs/vqa/ser/layoutxlm.yml

@@ -38,7 +38,7 @@ Optimizer:
     epochs: *epoch_num
     warmup_epoch: 2
   regularizer:
-    name: Const
+    name: L2
     factor: 0.00000
     
 PostProcess:

ppocr/optimizer/regularizer.py

@@ -29,39 +29,23 @@ class L1Decay(object):
 
     def __init__(self, factor=0.0):
         super(L1Decay, self).__init__()
-        self.regularization_coeff = factor
+        self.coeff = factor
 
     def __call__(self):
-        reg = paddle.regularizer.L1Decay(self.regularization_coeff)
+        reg = paddle.regularizer.L1Decay(self.coeff)
         return reg
 
 
 class L2Decay(object):
     """
-    L2 Weight Decay Regularization, which encourages the weights to be sparse.
+    L2 Weight Decay Regularization, which helps to prevent the model over-fitting.
     Args:
         factor(float): regularization coeff. Default:0.0.
     """
 
     def __init__(self, factor=0.0):
         super(L2Decay, self).__init__()
-        self.regularization_coeff = factor
+        self.coeff = factor
 
     def __call__(self):
-        reg = paddle.regularizer.L2Decay(self.regularization_coeff)
-        return reg
-
-
-class ConstDecay(object):
-    """
-    Const L2 Weight Decay Regularization, which encourages the weights to be sparse.
-    Args:
-        factor(float): regularization coeff. Default:0.0.
-    """
-
-    def __init__(self, factor=0.0):
-        super(ConstDecay, self).__init__()
-        self.regularization_coeff = factor
-
-    def __call__(self):
-        return self.regularization_coeff
+        return self.coeff