fix english doc, unittest, and remove useless alias of 2.0 lr_schedul…

…er (#27686)

* fix doc and unittest of 2.0 lr_scheduler

* fix doc of 2.0 lr_scheduler

* fix unittest

* fix english doc of lr_scheduler

* fix api name of lr scheduler

* fix api name of lr scheduler

python/paddle/__init__.py

@@ -237,13 +237,6 @@
 from .framework import load  #DEFINE_ALIAS
 from .framework import DataParallel  #DEFINE_ALIAS
 
-from .framework import NoamDecay  #DEFINE_ALIAS
-from .framework import PiecewiseDecay  #DEFINE_ALIAS
-from .framework import NaturalExpDecay  #DEFINE_ALIAS
-from .framework import ExponentialDecay  #DEFINE_ALIAS
-from .framework import InverseTimeDecay  #DEFINE_ALIAS
-from .framework import PolynomialDecay  #DEFINE_ALIAS
-from .framework import CosineDecay  #DEFINE_ALIAS
 from .framework import set_default_dtype  #DEFINE_ALIAS
 from .framework import get_default_dtype  #DEFINE_ALIAS
 

python/paddle/fluid/dygraph/checkpoint.py

@@ -164,7 +164,7 @@ def load_dygraph(model_path, **configs):
             state_dict = emb.state_dict()
             fluid.save_dygraph(state_dict, "paddle_dy")
 
-            scheduler = paddle.optimizer.lr_scheduler.NoamLR(	
+            scheduler = paddle.optimizer.lr.NoamDecay(	
                 d_model=0.01, warmup_steps=100, verbose=True)
             adam = paddle.optimizer.Adam(
                 learning_rate=scheduler,

python/paddle/fluid/executor.py

@@ -855,7 +855,7 @@ def close(self):
 
     def _run_parallel(self, program, scope, feed, fetch_list, fetch_var_name,
                       return_numpy, return_merged):
-        from paddle.optimizer.lr_scheduler import _LRScheduler
+        from paddle.optimizer.lr import LRScheduler
         exe = program._executor
         # TODO(zhenghuihuang): quantization uses Graph in CompiledProgram
         # instead of program. We will add support for checking Vars in Graph
@@ -901,7 +901,7 @@ def _run_parallel(self, program, scope, feed, fetch_list, fetch_var_name,
 
         if hasattr(program._program, 'lr_sheduler'):
             lr_sheduler = program._program.lr_sheduler
-            assert isinstance(lr_sheduler, _LRScheduler), "must be _LRScheduler"
+            assert isinstance(lr_sheduler, LRScheduler), "must be LRScheduler"
             lr_value = lr_sheduler()
             lr_var = program._program.global_block().vars[lr_sheduler._var_name]
             lr_tensor = _as_lodtensor(lr_value, core.CPUPlace(), lr_var.dtype)
@@ -1238,7 +1238,7 @@ def _run_impl(self, program, feed, fetch_list, feed_var_name,
 
     def _run_program(self, program, feed, fetch_list, feed_var_name,
                      fetch_var_name, scope, return_numpy, use_program_cache):
-        from paddle.optimizer.lr_scheduler import _LRScheduler
+        from paddle.optimizer.lr import LRScheduler
         if feed is None:
             feed = {}
         elif isinstance(feed, (list, tuple)):
@@ -1296,7 +1296,7 @@ def _run_program(self, program, feed, fetch_list, feed_var_name,
         self._feed_data(program, feed, feed_var_name, scope)
         if hasattr(program, 'lr_sheduler'):
             assert isinstance(program.lr_sheduler,
-                              _LRScheduler), "must be _LRScheduler"
+                              LRScheduler), "must be LRScheduler"
             lr_sheduler = program.lr_sheduler
             lr_value = lr_sheduler()
             lr_var = program.global_block().vars[lr_sheduler._var_name]

python/paddle/fluid/optimizer.py

@@ -70,15 +70,15 @@ def __init__(self,
                  grad_clip=None,
                  name=None):
         # Because of the loop import, so place it in the function body
-        from paddle.optimizer.lr_scheduler import _LRScheduler
+        from paddle.optimizer.lr import LRScheduler
         self._parameter_list = list(
             parameter_list) if parameter_list is not None else None
         self._name = name
         if framework.in_dygraph_mode():
             if not isinstance(learning_rate,
-                              (float, LearningRateDecay, _LRScheduler)):
+                              (float, LearningRateDecay, LRScheduler)):
                 raise TypeError(
-                    "learning rate should be float or _LRScheduler, got %s here"
+                    "learning rate should be float or LRScheduler, got %s here"
                     % type(learning_rate))
             if self._parameter_list is None:
                 raise AttributeError(
@@ -94,9 +94,9 @@ def __init__(self,
                         break
         else:
             if not isinstance(learning_rate,
-                              (float, framework.Variable, _LRScheduler)):
+                              (float, framework.Variable, LRScheduler)):
                 raise TypeError(
-                    "learning rate should be float or _LRScheduler, got %s here"
+                    "learning rate should be float or LRScheduler, got %s here"
                     % type(learning_rate))
 
         if grad_clip is not None:
@@ -147,13 +147,13 @@ def state_dict(self):
                     state_dict = adam.state_dict()
 
         '''
-        from paddle.optimizer.lr_scheduler import _LRScheduler
+        from paddle.optimizer.lr import LRScheduler
         state_dict = {}
         for k, v in self._accumulators.items():
             for para_name, var_tmp in v.items():
                 state_dict[var_tmp.name] = var_tmp
         # global step if use lr decay
-        if isinstance(self._learning_rate, _LRScheduler):
+        if isinstance(self._learning_rate, LRScheduler):
             state_dict["LR_Scheduler"] = self._learning_rate.state_dict()
             return state_dict
         if isinstance(self._learning_rate, LearningRateDecay):
@@ -193,7 +193,7 @@ def set_state_dict(self, state_dict):
                 state_dict = emb.state_dict()
                 fluid.save_dygraph(state_dict, "paddle_dy")
 
-                scheduler = paddle.optimizer.lr_scheduler.NoamLR(	
+                scheduler = paddle.optimizer.lr.NoamDecay(	
                     d_model=0.01, warmup_steps=100, verbose=True)
                 adam = paddle.optimizer.Adam(
                     learning_rate=scheduler,
@@ -203,8 +203,8 @@ def set_state_dict(self, state_dict):
 
                 para_state_dict, opti_state_dict = fluid.load_dygraph("paddle_dy")
         '''
-        from paddle.optimizer.lr_scheduler import _LRScheduler
-        if isinstance(self._learning_rate, _LRScheduler):
+        from paddle.optimizer.lr import LRScheduler
+        if isinstance(self._learning_rate, LRScheduler):
             self._learning_rate.set_dict(state_dict["LR_Scheduler"])
 
         if isinstance(self._learning_rate, LearningRateDecay):
@@ -269,8 +269,8 @@ def get_opti_var_name_list(self):
         return self._opti_name_list
 
     def _create_global_learning_rate(self):
-        from paddle.optimizer.lr_scheduler import _LRScheduler
-        if isinstance(self._learning_rate, _LRScheduler):
+        from paddle.optimizer.lr import LRScheduler
+        if isinstance(self._learning_rate, LRScheduler):
             lr_var = self._global_learning_rate()
             # only create global lr_var once
             if not isinstance(lr_var, framework.Variable):

python/paddle/fluid/tests/unittests/test_adam_op.py

@@ -455,8 +455,8 @@ def test_adam_op_with_state_dict(self):
         state_dict = adam.state_dict()
         adam.set_state_dict(state_dict)
 
-        #learning_rate is _LRScheduler
-        learning_rate = paddle.optimizer.CosineAnnealingLR(
+        #learning_rate is LRScheduler
+        learning_rate = paddle.optimizer.lr.CosineAnnealingDecay(
             learning_rate=0.1, T_max=10)
         adam = paddle.optimizer.Adam(
             learning_rate=learning_rate,

python/paddle/fluid/tests/unittests/test_directory_migration.py

@@ -43,14 +43,22 @@ def test_new_directory(self):
             'paddle.distributed.prepare_context', 'paddle.DataParallel',
             'paddle.jit', 'paddle.jit.TracedLayer', 'paddle.jit.to_static',
             'paddle.jit.ProgramTranslator', 'paddle.jit.TranslatedLayer',
-            'paddle.jit.save', 'paddle.jit.load', 'paddle.NoamDecay',
-            'paddle.PiecewiseDecay', 'paddle.NaturalExpDecay',
-            'paddle.ExponentialDecay', 'paddle.InverseTimeDecay',
-            'paddle.PolynomialDecay', 'paddle.CosineDecay',
-            'paddle.static.Executor', 'paddle.static.global_scope',
-            'paddle.static.scope_guard', 'paddle.static.append_backward',
-            'paddle.static.gradients', 'paddle.static.BuildStrategy',
-            'paddle.static.CompiledProgram', 'paddle.static.ExecutionStrategy',
+            'paddle.jit.save', 'paddle.jit.load',
+            'paddle.optimizer.lr.LRScheduler', 'paddle.optimizer.lr.NoamDecay',
+            'paddle.optimizer.lr.PiecewiseDecay',
+            'paddle.optimizer.lr.NaturalExpDecay',
+            'paddle.optimizer.lr.ExponentialDecay',
+            'paddle.optimizer.lr.InverseTimeDecay',
+            'paddle.optimizer.lr.PolynomialDecay',
+            'paddle.optimizer.lr.CosineAnnealingDecay',
+            'paddle.optimizer.lr.MultiStepDecay',
+            'paddle.optimizer.lr.StepDecay', 'paddle.optimizer.lr.LambdaDecay',
+            'paddle.optimizer.lr.ReduceOnPlateau',
+            'paddle.optimizer.lr.LinearWarmup', 'paddle.static.Executor',
+            'paddle.static.global_scope', 'paddle.static.scope_guard',
+            'paddle.static.append_backward', 'paddle.static.gradients',
+            'paddle.static.BuildStrategy', 'paddle.static.CompiledProgram',
+            'paddle.static.ExecutionStrategy',
             'paddle.static.default_main_program',
             'paddle.static.default_startup_program', 'paddle.static.Program',
             'paddle.static.name_scope', 'paddle.static.program_guard',