Merge pull request #2802 from reyoung/feature/fix_python_slow

Fix slow parsing a recursive depends topology

python/paddle/trainer_config_helpers/networks.py

@@ -1408,6 +1408,8 @@ def outputs(layers, *args):
     :return:
     """
 
+    traveled = set()
+
     def __dfs_travel__(layer,
                        predicate=lambda x: x.layer_type == LayerType.DATA):
         """
@@ -1419,6 +1421,11 @@ def outputs(layers, *args):
         :type layer: LayerOutput
         :return:
         """
+        if layer in traveled:
+            return []
+        else:
+            traveled.add(layer)
+
         assert isinstance(layer, LayerOutput), "layer is %s" % (layer)
         retv = []
         if layer.parents is not None:

python/paddle/trainer_config_helpers/tests/configs/file_list.sh

@@ -6,6 +6,7 @@ img_layers img_trans_layers util_layers simple_rnn_layers unused_layers test_cos
 test_rnn_group shared_fc shared_lstm shared_gru test_cost_layers_with_weight
 test_spp_layer test_bilinear_interp test_maxout test_bi_grumemory math_ops
 test_seq_concat_reshape test_pad test_smooth_l1 test_multiplex_layer
-test_prelu_layer test_row_conv test_detection_output_layer test_multibox_loss_layer)
+test_prelu_layer test_row_conv test_detection_output_layer test_multibox_loss_layer
+test_recursive_topology)
 
 export whole_configs=(test_split_datasource)

python/paddle/trainer_config_helpers/tests/configs/protostr/test_cost_layers_with_weight.protostr

@@ -131,6 +131,7 @@ input_layer_names: "weight"
 input_layer_names: "multi_class_label"
 output_layer_names: "__cost_0__"
 output_layer_names: "__mse_cost_0__"
+output_layer_names: "__nce_layer_0__"
 evaluators {
   name: "classification_error_evaluator"
   type: "classification_error"
@@ -154,6 +155,7 @@ sub_models {
   input_layer_names: "multi_class_label"
   output_layer_names: "__cost_0__"
   output_layer_names: "__mse_cost_0__"
+  output_layer_names: "__nce_layer_0__"
   evaluator_names: "classification_error_evaluator"
   is_recurrent_layer_group: false
 }

python/paddle/trainer_config_helpers/tests/configs/protostr/test_recursive_topology.protostr

+type: "nn"
+layers {
+  name: "data"
+  type: "data"
+  size: 100
+  active_type: ""
+}
+layers {
+  name: "__addto_0__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "data"
+  }
+  inputs {
+    input_layer_name: "data"
+  }
+}
+layers {
+  name: "__addto_1__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_0__"
+  }
+  inputs {
+    input_layer_name: "__addto_0__"
+  }
+}
+layers {
+  name: "__addto_2__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_1__"
+  }
+  inputs {
+    input_layer_name: "__addto_1__"
+  }
+}
+layers {
+  name: "__addto_3__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_2__"
+  }
+  inputs {
+    input_layer_name: "__addto_2__"
+  }
+}
+layers {
+  name: "__addto_4__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_3__"
+  }
+  inputs {
+    input_layer_name: "__addto_3__"
+  }
+}
+layers {
+  name: "__addto_5__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_4__"
+  }
+  inputs {
+    input_layer_name: "__addto_4__"
+  }
+}
+layers {
+  name: "__addto_6__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_5__"
+  }
+  inputs {
+    input_layer_name: "__addto_5__"
+  }
+}
+layers {
+  name: "__addto_7__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_6__"
+  }
+  inputs {
+    input_layer_name: "__addto_6__"
+  }
+}
+layers {
+  name: "__addto_8__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_7__"
+  }
+  inputs {
+    input_layer_name: "__addto_7__"
+  }
+}
+layers {
+  name: "__addto_9__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_8__"
+  }
+  inputs {
+    input_layer_name: "__addto_8__"
+  }
+}
+layers {
+  name: "__addto_10__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_9__"
+  }
+  inputs {
+    input_layer_name: "__addto_9__"
+  }
+}
+layers {
+  name: "__addto_11__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_10__"
+  }
+  inputs {
+    input_layer_name: "__addto_10__"
+  }
+}
+layers {
+  name: "__addto_12__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_11__"
+  }
+  inputs {
+    input_layer_name: "__addto_11__"
+  }
+}
+layers {
+  name: "__addto_13__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_12__"
+  }
+  inputs {
+    input_layer_name: "__addto_12__"
+  }
+}
+layers {
+  name: "__addto_14__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_13__"
+  }
+  inputs {
+    input_layer_name: "__addto_13__"
+  }
+}
+layers {
+  name: "__addto_15__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_14__"
+  }
+  inputs {
+    input_layer_name: "__addto_14__"
+  }
+}
+layers {
+  name: "__addto_16__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_15__"
+  }
+  inputs {
+    input_layer_name: "__addto_15__"
+  }
+}
+layers {
+  name: "__addto_17__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_16__"
+  }
+  inputs {
+    input_layer_name: "__addto_16__"
+  }
+}
+layers {
+  name: "__addto_18__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_17__"
+  }
+  inputs {
+    input_layer_name: "__addto_17__"
+  }
+}
+layers {
+  name: "__addto_19__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_18__"
+  }
+  inputs {
+    input_layer_name: "__addto_18__"
+  }
+}
+layers {
+  name: "__addto_20__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_19__"
+  }
+  inputs {
+    input_layer_name: "__addto_19__"
+  }
+}
+layers {
+  name: "__addto_21__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_20__"
+  }
+  inputs {
+    input_layer_name: "__addto_20__"
+  }
+}
+layers {
+  name: "__addto_22__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_21__"
+  }
+  inputs {
+    input_layer_name: "__addto_21__"
+  }
+}
+layers {
+  name: "__addto_23__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_22__"
+  }
+  inputs {
+    input_layer_name: "__addto_22__"
+  }
+}
+layers {
+  name: "__addto_24__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_23__"
+  }
+  inputs {
+    input_layer_name: "__addto_23__"
+  }
+}
+layers {
+  name: "__addto_25__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_24__"
+  }
+  inputs {
+    input_layer_name: "__addto_24__"
+  }
+}
+layers {
+  name: "__addto_26__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_25__"
+  }
+  inputs {
+    input_layer_name: "__addto_25__"
+  }
+}
+layers {
+  name: "__addto_27__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_26__"
+  }
+  inputs {
+    input_layer_name: "__addto_26__"
+  }
+}
+layers {
+  name: "__addto_28__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_27__"
+  }
+  inputs {
+    input_layer_name: "__addto_27__"
+  }
+}
+layers {
+  name: "__addto_29__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_28__"
+  }
+  inputs {
+    input_layer_name: "__addto_28__"
+  }
+}
+layers {
+  name: "__addto_30__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_29__"
+  }
+  inputs {
+    input_layer_name: "__addto_29__"
+  }
+}
+layers {
+  name: "__addto_31__"
+  type: "addto"
+  size: 100
+  active_type: ""
+  inputs {
+    input_layer_name: "__addto_30__"
+  }
+  inputs {
+    input_layer_name: "__addto_30__"
+  }
+}
+layers {
+  name: "__fc_layer_0__"
+  type: "fc"
+  size: 32
+  active_type: "relu"
+  inputs {
+    input_layer_name: "__addto_31__"
+    input_parameter_name: "___fc_layer_0__.w0"
+  }
+  bias_parameter_name: "___fc_layer_0__.wbias"
+}
+layers {
+  name: "__fc_layer_1__"
+  type: "fc"
+  size: 10
+  active_type: "softmax"
+  inputs {
+    input_layer_name: "__fc_layer_0__"
+    input_parameter_name: "___fc_layer_1__.w0"
+  }
+  bias_parameter_name: "___fc_layer_1__.wbias"
+}
+parameters {
+  name: "___fc_layer_0__.w0"
+  size: 3200
+  initial_mean: 0.0
+  initial_std: 0.1
+  dims: 100
+  dims: 32
+  initial_strategy: 0
+  initial_smart: true
+}
+parameters {
+  name: "___fc_layer_0__.wbias"
+  size: 32
+  initial_mean: 0.0
+  initial_std: 0.0
+  dims: 1
+  dims: 32
+  initial_strategy: 0
+  initial_smart: false
+}
+parameters {
+  name: "___fc_layer_1__.w0"
+  size: 320
+  initial_mean: 0.0
+  initial_std: 0.176776695297
+  dims: 32
+  dims: 10
+  initial_strategy: 0
+  initial_smart: true
+}
+parameters {
+  name: "___fc_layer_1__.wbias"
+  size: 10
+  initial_mean: 0.0
+  initial_std: 0.0
+  dims: 1
+  dims: 10
+  initial_strategy: 0
+  initial_smart: false
+}
+input_layer_names: "data"
+output_layer_names: "__fc_layer_1__"
+sub_models {
+  name: "root"
+  layer_names: "data"
+  layer_names: "__addto_0__"
+  layer_names: "__addto_1__"
+  layer_names: "__addto_2__"
+  layer_names: "__addto_3__"
+  layer_names: "__addto_4__"
+  layer_names: "__addto_5__"
+  layer_names: "__addto_6__"
+  layer_names: "__addto_7__"
+  layer_names: "__addto_8__"
+  layer_names: "__addto_9__"
+  layer_names: "__addto_10__"
+  layer_names: "__addto_11__"
+  layer_names: "__addto_12__"
+  layer_names: "__addto_13__"
+  layer_names: "__addto_14__"
+  layer_names: "__addto_15__"
+  layer_names: "__addto_16__"
+  layer_names: "__addto_17__"
+  layer_names: "__addto_18__"
+  layer_names: "__addto_19__"
+  layer_names: "__addto_20__"
+  layer_names: "__addto_21__"
+  layer_names: "__addto_22__"
+  layer_names: "__addto_23__"
+  layer_names: "__addto_24__"
+  layer_names: "__addto_25__"
+  layer_names: "__addto_26__"
+  layer_names: "__addto_27__"
+  layer_names: "__addto_28__"
+  layer_names: "__addto_29__"
+  layer_names: "__addto_30__"
+  layer_names: "__addto_31__"
+  layer_names: "__fc_layer_0__"
+  layer_names: "__fc_layer_1__"
+  input_layer_names: "data"
+  output_layer_names: "__fc_layer_1__"
+  is_recurrent_layer_group: false
+}
+

python/paddle/trainer_config_helpers/tests/configs/test_recursive_topology.py

+from paddle.trainer_config_helpers import *
+
+settings(batch_size=1000, learning_rate=1e-5)
+
+din = data_layer(name='data', size=100)
+
+enc = din
+for i in range(32):
+    enc = addto_layer([enc, enc])
+
+pred = fc_layer(
+    input=fc_layer(
+        input=enc, size=32, act=ReluActivation()),
+    size=10,
+    act=SoftmaxActivation())
+outputs(pred)