fix API optimizer get_lr to support static graph mode (#28681)

* fix doc of save/load

* fix API optimizer get_lr

* fix API optimizer get_lr

python/paddle/optimizer/optimizer.py

@@ -349,52 +349,61 @@ class Optimizer(object):
                 },
                 stop_gradient=True)
 
-    @framework.dygraph_only
     def get_lr(self):
         """
-        :api_attr: imperative
+        Get current learning rate of optimizer. 
-        
+        If 'LRScheduler' is not used, the return value is all the same.
-        Get current step learning rate. The return value is all the same When LRScheduler is not used,
+        If 'LRScheduler' is used, the return value is the current scheduled learing rete.
-        otherwise return the current step learning rate.
-
 
         Returns:
-            float: The learning rate of the current step.
+            float: The current learning rate of optimizer.
 
         Examples:
             .. code-block:: python
 
-                import numpy as np
+                # train on default dynamic graph mode
                 import paddle
-                # example1: LRScheduler is not used, return value is all the same
+                import numpy as np
-                emb = paddle.nn.Embedding(10, 10)
+                emb = paddle.nn.Embedding(10, 3)
-                adam = paddle.optimizer.Adam(0.001, parameters = emb.parameters())
+
-                lr = adam.get_lr()
+                ## example1: LRScheduler is not used, return the same value is all the same
-                print(lr) # 0.001
+                adam = paddle.optimizer.Adam(0.01, parameters = emb.parameters())
+                for batch in range(10):
+                    input = paddle.randint(low=0, high=5, shape=[5])
+                    out = emb(input)
+                    out.backward()
+                    print("Learning rate of step{}: {}".format(batch, adam.get_lr())) # 0.01
+                    adam.step()
 
-                # example2: PiecewiseDecay is used, return the scheduled learning rate
+                ## example2: StepDecay is used, return the scheduled learning rate
-                inp = np.random.uniform(-0.1, 0.1, [10, 10]).astype("float32")
+                scheduler = paddle.optimizer.lr.StepDecay(learning_rate=0.5, step_size=2, gamma=0.1)
-                linear = paddle.nn.Linear(10, 10)
+                adam = paddle.optimizer.Adam(scheduler, parameters = emb.parameters())
-                inp = paddle.to_tensor(inp)
+                for batch in range(10):
-                out = linear(inp)
+                    input = paddle.randint(low=0, high=5, shape=[5])
-                loss = paddle.mean(out)
+                    out = emb(input)
-                
+                    out.backward()
-                bd = [2, 4, 6, 8]
+                    print("Learning rate of step{}: {}".format(batch, adam.get_lr())) # 0.5->0.05...
-                value = [0.2, 0.4, 0.6, 0.8, 1.0]
-                scheduler = paddle.optimizer.lr.PiecewiseDecay(bd, value, 0)
-                adam = paddle.optimizer.Adam(scheduler,
-                                       parameters=linear.parameters())
-
-                # first step: learning rate is 0.2
-                np.allclose(adam.get_lr(), 0.2, rtol=1e-06, atol=0.0) # True
-
-                # learning rate for different steps
-                ret = [0.2, 0.2, 0.4, 0.4, 0.6, 0.6, 0.8, 0.8, 1.0, 1.0, 1.0, 1.0]
-                for i in range(12):
                     adam.step()
-                    lr = adam.get_lr()
                     scheduler.step()
-                    np.allclose(lr, ret[i], rtol=1e-06, atol=0.0) # True
+
+                # train on static graph mode
+                paddle.enable_static()
+                main_prog = paddle.static.Program()
+                start_prog = paddle.static.Program()
+                with paddle.static.program_guard(main_prog, start_prog):
+                    x = paddle.static.data(name='x', shape=[None, 10])
+                    z = paddle.static.nn.fc(x, 100)
+                    loss = paddle.mean(z)
+                    scheduler = paddle.optimizer.lr.StepDecay(learning_rate=0.5, step_size=2, gamma=0.1)
+                    adam = paddle.optimizer.Adam(learning_rate=scheduler)
+                    adam.minimize(loss)
+
+                exe = paddle.static.Executor()
+                exe.run(start_prog)
+                for batch in range(10):
+                    print("Learning rate of step{}: {}", adam.get_lr())     # 0.5->0.05->0.005...
+                    out = exe.run(main_prog, feed={'x': np.random.randn(3, 10).astype('float32')})
+                    scheduler.step()
 
         """
         if isinstance(self._learning_rate, float):