[remove fluid.layers.cross_entropy] remove unit tests (part 3) (#48918)

* replace cross_entropy in python/paddle/fluid/tests/unittests/test_[o-z]*.py plus test_dist_transpiler.py

* fix test_prune

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

@@ -747,7 +747,9 @@ class TestDistLookupTableBase(TranspilerTest):
         )
 
         label = fluid.layers.data(name='label', shape=[1], dtype='int64')
-        cost = fluid.layers.cross_entropy(input=predict, label=label)
+        cost = paddle.nn.functional.cross_entropy(
+            input=predict, label=label, reduction='none', use_softmax=False
+        )
         avg_cost = paddle.mean(cost)
         optimizer = fluid.optimizer.Adam(learning_rate=0.003)
         optimizer.minimize(avg_cost)

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

@@ -1168,7 +1168,12 @@ class TestRecomputeOptimizer(unittest.TestCase):
             prediction = fluid.layers.fc(
                 input=[drop_res], size=2, act='softmax'
             )
-            cost = fluid.layers.cross_entropy(input=prediction, label=input_y)
+            cost = paddle.nn.functional.cross_entropy(
+                input=prediction,
+                label=input_y,
+                reduction='none',
+                use_softmax=False,
+            )
             sum_cost = paddle.mean(cost)
             return drop_res, prediction, sum_cost
 
@@ -1225,7 +1230,12 @@ class TestRecomputeOptimizerCUDA(unittest.TestCase):
             prediction = fluid.layers.fc(
                 input=[drop_res], size=2, act='softmax'
             )
-            cost = fluid.layers.cross_entropy(input=prediction, label=input_y)
+            cost = paddle.nn.functional.cross_entropy(
+                input=prediction,
+                label=input_y,
+                reduction='none',
+                use_softmax=False,
+            )
             sum_cost = paddle.mean(cost)
             return drop_res, prediction, sum_cost
 

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

@@ -75,7 +75,9 @@ def static(
 
         def fn_1(opt, avg_loss=None, pred=None, label=None):
             if avg_loss is None:
-                loss = layers.cross_entropy(input=pred, label=label)
+                loss = paddle.nn.functional.cross_entropy(
+                    input=pred, label=label, reduction='none', use_softmax=False
+                )
                 avg_loss = paddle.mean(loss, name='mean_cross_entropy_loss')
             opt.minimize(avg_loss)
             return avg_loss
@@ -106,7 +108,12 @@ def static(
                 lambda: fn_2(sgd, None, prediction, label),
             )
         else:
-            loss_1 = layers.cross_entropy(input=prediction, label=label)
+            loss_1 = paddle.nn.functional.cross_entropy(
+                input=prediction,
+                label=label,
+                reduction='none',
+                use_softmax=False,
+            )
             avg_loss_1 = paddle.mean(loss_1)
             loss_2 = paddle.nn.functional.softmax_with_cross_entropy(
                 logits=prediction, label=label
@@ -188,7 +195,9 @@ def dynamic(train_data, use_cuda=False, use_parallel_exe=False):
             hidden, prediction = dy_layer(var_input)
 
             if epoch % 2 == 0:
-                cross_entropy_loss = layers.cross_entropy(prediction, var_label)
+                cross_entropy_loss = paddle.nn.functional.cross_entropy(
+                    prediction, var_label, reduction='none', use_softmax=False
+                )
                 loss = paddle.mean(cross_entropy_loss)
                 loss.backward()
                 adam.minimize(loss)

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

@@ -82,7 +82,9 @@ class TestMNISTDryRun(TestBase):
         for _ in range(10):
             hidden = fluid.layers.fc(input=img, size=200, act='tanh')
         prediction = fluid.layers.fc(input=hidden, size=10, act='softmax')
-        loss = fluid.layers.cross_entropy(input=prediction, label=label)
+        loss = paddle.nn.functional.cross_entropy(
+            input=prediction, label=label, reduction='none', use_softmax=False
+        )
         avg_loss = paddle.mean(loss)
         fluid.optimizer.Adam().minimize(avg_loss)
         return avg_loss

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

@@ -60,7 +60,9 @@ class TestFetchAndFeed(unittest.TestCase):
             )
             label = fluid.layers.data(name='label', shape=[1], dtype='int64')
             out = Lenet(data, class_dim=102)
-            loss = fluid.layers.cross_entropy(input=out, label=label)
+            loss = paddle.nn.functional.cross_entropy(
+                input=out, label=label, reduction='none', use_softmax=False
+            )
             loss = paddle.mean(loss)
             opt = fluid.optimizer.Momentum(
                 learning_rate=0.1,

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

@@ -37,7 +37,9 @@ def simple_fc_net(use_feed):
             ),
         )
     prediction = fluid.layers.fc(hidden, size=10, act='softmax')
-    loss = fluid.layers.cross_entropy(input=prediction, label=label)
+    loss = paddle.nn.functional.cross_entropy(
+        input=prediction, label=label, reduction='none', use_softmax=False
+    )
     loss = paddle.mean(loss)
     return loss
 
@@ -62,7 +64,9 @@ def fc_with_batchnorm(use_feed):
     with fluid.name_scope("fc_layer"):
         prediction = fluid.layers.fc(hidden, size=10, act='softmax')
     with fluid.name_scope("loss"):
-        loss = fluid.layers.cross_entropy(input=prediction, label=label)
+        loss = paddle.nn.functional.cross_entropy(
+            input=prediction, label=label, reduction='none', use_softmax=False
+        )
         loss = paddle.mean(loss)
     return loss
 

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

@@ -57,7 +57,9 @@ class TestProfiler(unittest.TestCase):
             hidden2 = fluid.layers.fc(input=hidden_n, size=64, act='relu')
             predict = fluid.layers.fc(input=hidden2, size=10, act='softmax')
             label = fluid.layers.data(name='y', shape=[1], dtype='int64')
-            cost = fluid.layers.cross_entropy(input=predict, label=label)
+            cost = paddle.nn.functional.cross_entropy(
+                input=predict, label=label, reduction='none', use_softmax=False
+            )
             avg_cost = paddle.mean(cost)
             batch_size = paddle.tensor.create_tensor(dtype='int64')
             batch_acc = paddle.static.accuracy(

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

@@ -54,7 +54,9 @@ def lstm_net(use_feed):
     lstm_max_tanh = paddle.tanh(lstm_max)
     fc1 = fluid.layers.fc(input=lstm_max_tanh, size=hid_dim2, act='tanh')
     prediction = fluid.layers.fc(input=fc1, size=class_dim, act='softmax')
-    cost = fluid.layers.cross_entropy(input=prediction, label=label)
+    cost = paddle.nn.functional.cross_entropy(
+        input=prediction, label=label, reduction='none', use_softmax=False
+    )
     avg_cost = paddle.mean(x=cost)
     return avg_cost
 
@@ -74,7 +76,9 @@ def simple_fc_net_with_accuracy(use_feed):
             ),
         )
     prediction = fluid.layers.fc(hidden, size=10, act='softmax')
-    loss = fluid.layers.cross_entropy(input=prediction, label=label)
+    loss = paddle.nn.functional.cross_entropy(
+        input=prediction, label=label, reduction='none', use_softmax=False
+    )
     loss = paddle.mean(loss)
     accuracy_out = paddle.static.accuracy(input=prediction, label=label, k=5)
     return loss
@@ -87,7 +91,9 @@ def cond_net(use_feed=None):
 
     def loss1(pred, label):
         x = fluid.layers.data(name="x", shape=[4], dtype='float32')
-        loss = fluid.layers.cross_entropy(input=pred, label=label)
+        loss = paddle.nn.functional.cross_entropy(
+            input=pred, label=label, reduction='none', use_softmax=False
+        )
         avg_loss = paddle.mean(loss, name='mean_cross_entropy_loss')
         return avg_loss
 
@@ -114,7 +120,9 @@ def optimization_in_cond_net(with_optimize=False):
 
     def loss1(opt, pred, label, with_optimize):
         x = fluid.layers.data(name="x", shape=[4], dtype='float32')
-        loss = fluid.layers.cross_entropy(input=pred, label=label)
+        loss = paddle.nn.functional.cross_entropy(
+            input=pred, label=label, reduction='none', use_softmax=False
+        )
         avg_loss = paddle.mean(loss, name='mean_cross_entropy_loss')
         if with_optimize:
             opt.minimize(avg_loss)

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

@@ -28,7 +28,9 @@ class TestPrune(unittest.TestCase):
         x = fluid.layers.data(name='x', shape=[2], dtype='float32')
         label = fluid.layers.data(name="label", shape=[1], dtype="int64")
         y = fluid.layers.fc(input=[x], size=2, act="softmax")
-        loss = fluid.layers.cross_entropy(input=y, label=label)
+        loss = paddle.nn.functional.cross_entropy(
+            input=y, label=label, reduction='none', use_softmax=False
+        )
         loss = paddle.mean(x=loss)
         return x, y, label, loss
 
@@ -45,7 +47,7 @@ class TestPrune(unittest.TestCase):
                 "mul",
                 "elementwise_add",
                 "softmax",
-                "cross_entropy2",
+                "softmax_with_cross_entropy",
                 "reduce_mean",
             ],
         )
@@ -55,7 +57,7 @@ class TestPrune(unittest.TestCase):
         self.assertEqual(len(pruned_program.global_block().ops), 2)
         self.assertEqual(
             [op.type for op in pruned_program.global_block().ops],
-            ["cross_entropy2", "reduce_mean"],
+            ["softmax_with_cross_entropy", "reduce_mean"],
         )
 
     def test_prune(self):
@@ -71,7 +73,7 @@ class TestPrune(unittest.TestCase):
                 "mul",
                 "elementwise_add",
                 "softmax",
-                "cross_entropy2",
+                "softmax_with_cross_entropy",
                 "reduce_mean",
             ],
         )
@@ -83,7 +85,7 @@ class TestPrune(unittest.TestCase):
                 "mul",
                 "elementwise_add",
                 "softmax",
-                "cross_entropy2",
+                "softmax_with_cross_entropy",
                 "reduce_mean",
             ],
         )
@@ -101,7 +103,7 @@ class TestPrune(unittest.TestCase):
                 "mul",
                 "elementwise_add",
                 "softmax",
-                "cross_entropy2",
+                "softmax_with_cross_entropy",
                 "reduce_mean",
             ],
         )
@@ -113,7 +115,7 @@ class TestPrune(unittest.TestCase):
                 "mul",
                 "elementwise_add",
                 "softmax",
-                "cross_entropy2",
+                "softmax_with_cross_entropy",
                 "reduce_mean",
             ],
         )
@@ -131,7 +133,7 @@ class TestPrune(unittest.TestCase):
                 "mul",
                 "elementwise_add",
                 "softmax",
-                "cross_entropy2",
+                "softmax_with_cross_entropy",
                 "reduce_mean",
             ],
         )
@@ -170,9 +172,13 @@ class TestExecutorRunAutoPrune(unittest.TestCase):
         y = fluid.layers.fc(
             input=[x], size=2, act="softmax", param_attr=w_param_attrs
         )
-        loss1 = fluid.layers.cross_entropy(input=y, label=label)
+        loss1 = paddle.nn.functional.cross_entropy(
+            input=y, label=label, reduction='none', use_softmax=False
+        )
         loss1 = paddle.mean(x=loss1)
-        loss2 = fluid.layers.cross_entropy(input=y, label=label)
+        loss2 = paddle.nn.functional.cross_entropy(
+            input=y, label=label, reduction='none', use_softmax=False
+        )
         loss2 = paddle.mean(x=loss2)
         loss1.persistable = True
         loss2.persistable = True
@@ -200,9 +206,13 @@ class TestExecutorRunAutoPrune(unittest.TestCase):
         y2 = fluid.layers.fc(
             input=[x2], size=2, act="softmax", param_attr=w2_param_attrs
         )
-        loss1 = fluid.layers.cross_entropy(input=y1, label=label)
+        loss1 = paddle.nn.functional.cross_entropy(
+            input=y1, label=label, reduction='none', use_softmax=False
+        )
         loss1 = paddle.mean(x=loss1)
-        loss2 = fluid.layers.cross_entropy(input=y2, label=label)
+        loss2 = paddle.nn.functional.cross_entropy(
+            input=y2, label=label, reduction='none', use_softmax=False
+        )
         loss2 = paddle.mean(x=loss2)
         return (
             x1,

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

@@ -104,7 +104,9 @@ def simple_fc_net(img, label, use_py_func_op):
 
     prediction = fluid.layers.fc(hidden, size=10, act='softmax')
     if not use_py_func_op:
-        loss = fluid.layers.cross_entropy(input=prediction, label=label)
+        loss = paddle.nn.functional.cross_entropy(
+            input=prediction, label=label, reduction='none', use_softmax=False
+        )
     else:
         loss = (
             fluid.default_main_program()

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

@@ -141,7 +141,9 @@ def bow_net(
     fc_1 = fluid.layers.fc(input=bow_tanh, size=hid_dim, act="tanh")
     fc_2 = fluid.layers.fc(input=fc_1, size=hid_dim2, act="tanh")
     prediction = fluid.layers.fc(input=[fc_2], size=class_dim, act="softmax")
-    cost = fluid.layers.cross_entropy(input=prediction, label=label)
+    cost = paddle.nn.functional.cross_entropy(
+        input=prediction, label=label, reduction='none', use_softmax=False
+    )
     avg_cost = paddle.mean(x=cost)
     return avg_cost
 

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

@@ -47,7 +47,9 @@ def bow_net(
     fc_1 = fluid.layers.fc(input=bow_tanh, size=hid_dim, act="tanh")
     fc_2 = fluid.layers.fc(input=fc_1, size=hid_dim2, act="tanh")
     prediction = fluid.layers.fc(input=[fc_2], size=class_dim, act="softmax")
-    cost = fluid.layers.cross_entropy(input=prediction, label=label)
+    cost = paddle.nn.functional.cross_entropy(
+        input=prediction, label=label, reduction='none', use_softmax=False
+    )
     avg_cost = paddle.mean(x=cost)
 
     return avg_cost

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

@@ -296,7 +296,9 @@ class PolicyGradient:
         self.reward = paddle.static.py_func(
             func=reward_func, x=[action, length], out=reward
         )
-        neg_log_prob = layers.cross_entropy(act_prob, action)
+        neg_log_prob = paddle.nn.functional.cross_entropy(
+            act_prob, action, reduction='none', use_softmax=False
+        )
         cost = neg_log_prob * reward
         cost = (
             (paddle.sum(cost) / paddle.sum(length))
@@ -383,7 +385,13 @@ class MLE:
         self.lr = lr
 
     def learn(self, probs, label, weight=None, length=None):
-        loss = layers.cross_entropy(input=probs, label=label, soft_label=False)
+        loss = paddle.nn.functional.cross_entropy(
+            input=probs,
+            label=label,
+            soft_label=False,
+            reduction='none',
+            use_softmax=False,
+        )
         max_seq_len = paddle.shape(probs)[1]
         mask = layers.sequence_mask(length, maxlen=max_seq_len, dtype="float32")
         loss = loss * mask

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

@@ -27,7 +27,9 @@ def test_trainable():
     feature = fluid.layers.fc(
         input=x, size=10, param_attr=fluid.ParamAttr(trainable=False)
     )
-    loss = fluid.layers.cross_entropy(input=feature, label=label)
+    loss = paddle.nn.functional.cross_entropy(
+        input=feature, label=label, reduction='none', use_softmax=False
+    )
     loss = paddle.mean(loss)
     return loss
 

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

@@ -63,7 +63,9 @@ def bow_net(
     fc_1 = fluid.layers.fc(input=bow_tanh, size=hid_dim, act="tanh")
     fc_2 = fluid.layers.fc(input=fc_1, size=hid_dim2, act="tanh")
     prediction = fluid.layers.fc(input=[fc_2], size=class_dim, act="softmax")
-    cost = fluid.layers.cross_entropy(input=prediction, label=label)
+    cost = paddle.nn.functional.cross_entropy(
+        input=prediction, label=label, reduction='none', use_softmax=False
+    )
     avg_cost = paddle.mean(x=cost)
 
     return avg_cost