Feature/metrics (#9791)

* "add metrics"

* "add fluid metrics"

* "add import guards"

* "show warnings"

* "add demo"

* "fix ci"

* "add some details"

* "fix cci"

* "add demo Python"

* "add metrics"

benchmark/fluid/mnist.py

@@ -139,9 +139,6 @@ def run_benchmark(model, args):
 
     # inference program
     inference_program = fluid.default_main_program().clone()
-    with fluid.program_guard(inference_program):
-        inference_program = fluid.io.get_inference_program(
-            target_vars=[batch_acc, batch_size_tensor])
 
     # Optimization
     opt = fluid.optimizer.AdamOptimizer(
@@ -161,7 +158,7 @@ def run_benchmark(model, args):
     train_reader = paddle.batch(
         paddle.dataset.mnist.train(), batch_size=args.batch_size)
 
-    accuracy = fluid.average.WeightedAverage()
+    accuracy = fluid.metrics.Accuracy()
     iters, num_samples, start_time = 0, 0, time.time()
     for pass_id in range(args.pass_num):
         accuracy.reset()
@@ -184,7 +181,7 @@ def run_benchmark(model, args):
                       "label": y_data},
                 fetch_list=[avg_cost, batch_acc, batch_size_tensor]
             )  # The accuracy is the accumulation of batches, but not the current batch.
-            accuracy.add(value=outs[1], weight=outs[2])
+            accuracy.update(value=outs[1], weight=outs[2])
             iters += 1
             num_samples += len(y_data)
             loss = np.array(outs[0])

python/paddle/fluid/__init__.py

@@ -29,6 +29,7 @@ import optimizer
 import backward
 import regularizer
 import average
+import metrics
 from param_attr import ParamAttr, WeightNormParamAttr
 from data_feeder import DataFeeder
 from core import LoDTensor, CPUPlace, CUDAPlace, CUDAPinnedPlace

python/paddle/fluid/average.py

@@ -13,6 +13,7 @@
 # limitations under the License.
 
 import numpy as np
+import warnings
 """
     Class of all kinds of Average.
 
@@ -22,6 +23,8 @@ import numpy as np
     wrappers of Python functions.
 """
 
+__all__ = ["WeightedAverage"]
+
 
 def _is_number_(var):
     return isinstance(var, int) or isinstance(var, float) or (isinstance(
@@ -34,6 +37,9 @@ def _is_number_or_matrix_(var):
 
 class WeightedAverage(object):
     def __init__(self):
+        warnings.warn(
+            "The %s is deprecated, please use fluid.metrics.Accuracy instead." %
+            (self.__class__.__name__), Warning)
         self.reset()
 
     def reset(self):

python/paddle/fluid/evaluator.py

@@ -12,6 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import warnings
 import numpy as np
 
 import layers
@@ -59,6 +60,9 @@ class Evaluator(object):
     """
 
     def __init__(self, name, **kwargs):
+        warnings.warn(
+            "The %s is deprecated, because maintain a modified program inside evaluator cause bug easily, please use fluid.metrics.%s instead."
+            % (self.__class__.__name__, self.__class__.__name__), Warning)
         self.states = []
         self.metrics = []
         self.helper = LayerHelper(name, **kwargs)

python/paddle/fluid/layers/metric.py

@@ -15,12 +15,13 @@
 All layers just related to metric.
 """
 
+import warnings
 from ..layer_helper import LayerHelper
 from ..initializer import Normal, Constant
 from ..framework import Variable
 from ..param_attr import ParamAttr
 
-__all__ = ['accuracy']
+__all__ = ['accuracy', 'auc']
 
 
 def accuracy(input, label, k=1, correct=None, total=None):
@@ -55,3 +56,37 @@ def accuracy(input, label, k=1, correct=None, total=None):
             "Total": [total],
         })
     return acc_out
+
+
+def auc(input, label, curve='ROC', num_thresholds=200):
+    warnings.warn(
+        "This interface not recommended, fluid.layers.auc compute the auc at every minibatch, \
+        but can not aggregate them and get the pass AUC, because pass \
+        auc can not be averaged with weighted from the minibatch auc value. \
+        Please use fluid.metrics.Auc, it can compute the auc value via Python natively, \
+        which can get every minibatch and every pass auc value.", Warning)
+    helper = LayerHelper("auc", **locals())
+    topk_out = helper.create_tmp_variable(dtype=input.dtype)
+    topk_indices = helper.create_tmp_variable(dtype="int64")
+    helper.append_op(
+        type="top_k",
+        inputs={"X": [input]},
+        outputs={"Out": [topk_out],
+                 "Indices": [topk_indices]},
+        attrs={"k": k})
+    auc_out = helper.create_tmp_variable(dtype="float32")
+    if correct is None:
+        correct = helper.create_tmp_variable(dtype="int64")
+    if total is None:
+        total = helper.create_tmp_variable(dtype="int64")
+    helper.append_op(
+        type="accuracy",
+        inputs={
+            "Out": [topk_out],
+            "Indices": [topk_indices],
+            "Label": [label]
+        },
+        attrs={"curve": curve,
+               "num_thresholds": num_thresholds},
+        outputs={"AUC": [auc_out], })
+    return auc_out