LRScheduler with update_on_kvstore=False

apache · Mar 9, 2019 · 38671ab · 38671ab
1 parent ed83071
commit 38671ab
Show file tree

Hide file tree

Showing 3 changed files with 25 additions and 14 deletions.
diff --git a/python/mxnet/gluon/trainer.py b/python/mxnet/gluon/trainer.py
@@ -241,10 +241,10 @@ def _init_kvstore(self):
                 kvstore.set_optimizer(self._optimizer)
             self._kvstore = kvstore
             self._update_on_kvstore = update_on_kvstore
-            if self._optimizer.lr_scheduler and not self._update_on_kvstore:
-                raise ValueError("update_on_kvstore=False does not support " \
-                                 "optimizer with LRScheduler. Please " \
-                                 "consider setting learning rate manually.")
+            #if self._optimizer.lr_scheduler and not self._update_on_kvstore:
+            #    raise ValueError("update_on_kvstore=False does not support " \
+            #                     "optimizer with LRScheduler. Please " \
+            #                     "consider setting learning rate manually.")
         else:
             self._kvstore = None
             self._update_on_kvstore = None

diff --git a/python/mxnet/optimizer/optimizer.py b/python/mxnet/optimizer/optimizer.py
@@ -106,7 +106,8 @@ def __init__(self, rescale_grad=1., param_idx2name=None, wd=0.,
         self.wd_mult = {}
         self.begin_num_update = begin_num_update
         self.num_update = begin_num_update
-        self._index_update_count = {}
+        self._all_index_update_counts = {0 : {}}
+        self._index_update_count = self._all_index_update_counts[0]
         self.clip_gradient = clip_gradient
         self.multi_precision = multi_precision
         self.aggregate_num = 0
@@ -380,6 +381,11 @@ def set_wd_mult(self, args_wd_mult):
                     self.wd_mult[name] = float(attr[name]['__wd_mult__'])
         self.wd_mult.update(args_wd_mult)
 
+    def set_current_context(self, ctx):
+        if ctx not in self._all_index_update_counts:
+            self._all_index_update_counts[ctx] = {}
+        self._index_update_count = self._all_index_update_counts[ctx]
+
     def _update_count(self, index):
         """Updates num_update.
 
@@ -1623,6 +1629,8 @@ def __call__(self, index, grad, weight):
             indices = index
             grads = grad
             weights = weight
+        if weights:
+            self.optimizer.set_current_context(weights[0].context.device_id)
         for i, idx in enumerate(indices):
             # convert ctypes.char_p.value back to python str if needed
             if isinstance(idx, bytes):

diff --git a/tests/python/unittest/test_gluon_trainer.py b/tests/python/unittest/test_gluon_trainer.py
@@ -272,19 +272,22 @@ def test_trainer_lr_sched():
             lr *= factor
     mx.nd.waitall()
 
-@with_seed()
-def test_trainer_invalid_lr_sched():
+    # Update on kvstore = False
     x = gluon.Parameter('x', shape=(10,))
     x.initialize(ctx=[mx.cpu(0), mx.cpu(1)], init='zeros')
     freq = 2
     factor = 0.1
     lr = 1
     lr_sched = mx.lr_scheduler.FactorScheduler(freq, factor=factor, base_lr=lr)
-    invalid_trainer = gluon.Trainer([x], 'sgd', {'learning_rate': lr, 'lr_scheduler': lr_sched},
-                                    update_on_kvstore=False)
-    with mx.autograd.record():
-        for w in x.list_data():
-            y = w + 1
-            y.backward()
-    assert_raises(ValueError, invalid_trainer.step, 1)
+    trainer = gluon.Trainer([x], 'sgd', {'learning_rate': lr, 'lr_scheduler': lr_sched},
+                            update_on_kvstore=False)
+    for i in range(10):
+        with mx.autograd.record():
+            for w in x.list_data():
+                y = w + 1
+                y.backward()
+        trainer.step(1)
+        if i % freq == 0:
+            assert trainer.learning_rate == lr, (lr, trainer.learning_rate, i)
+            lr *= factor
     mx.nd.waitall()