Merge pull request #86 from DesmonDay/add_vdl_logwriter

Function add: add the VisualDL log writer for pgl

examples/erniesage/learner.py

@@ -17,6 +17,7 @@ role = os.getenv("TRAINING_ROLE", "TRAINER")
 
 import numpy as np
 from pgl.utils.logger import log
+from pgl.utils.log_writer import LogWriter
 import paddle.fluid as F
 import paddle.fluid.layers as L
 from paddle.fluid.incubate.fleet.parameter_server.distribute_transpiler import StrategyFactory
@@ -25,7 +26,6 @@ from paddle.fluid.transpiler.distribute_transpiler import DistributeTranspilerCo
 from paddle.fluid.incubate.fleet.collective import fleet as cfleet
 from paddle.fluid.incubate.fleet.parameter_server.distribute_transpiler import fleet as tfleet
 import paddle.fluid.incubate.fleet.base.role_maker as role_maker
-from tensorboardX import SummaryWriter
 from paddle.fluid.transpiler.distribute_transpiler import DistributedMode
 from paddle.fluid.incubate.fleet.parameter_server.distribute_transpiler.distributed_strategy import TrainerRuntimeConfig
 
@@ -77,7 +77,7 @@ class Learner(object):
         start = time.time()
         trainer_id = int(os.getenv("PADDLE_TRAINER_ID", "0"))
         if trainer_id == 0:
-            writer = SummaryWriter(os.path.join(self.config.output_path, "train_history"))
+            writer = LogWriter(os.path.join(self.config.output_path, "train_history"))
 
         for epoch_idx in range(self.config.epoch):
             for idx, batch_feed_dict in enumerate(self.model.data_loader()):

ogb_examples/graphproppred/mol/monitor/train_monitor.py

@@ -19,10 +19,10 @@ import os
 from datetime import datetime
 import logging
 from collections import defaultdict
-from tensorboardX import SummaryWriter
 
 import paddle.fluid as F
 from pgl.utils.logger import log
+from pgl.utils.log_writer import LogWriter
 
 
 def multi_device(reader, dev_count):
@@ -79,10 +79,10 @@ def train_and_evaluate(exe,
     global_step = 0
 
     timestamp = datetime.now().strftime("%Hh%Mm%Ss")
-    log_path = os.path.join(args.log_dir, "tensorboard_log_%s" % timestamp)
+    log_path = os.path.join(args.log_dir, "log_%s" % timestamp)
     _create_if_not_exist(log_path)
 
-    writer = SummaryWriter(log_path)
+    writer = LogWriter(log_path)
 
     best_valid_score = 0.0
     for e in range(args.epoch):
@@ -99,7 +99,7 @@ def train_and_evaluate(exe,
             ret = model.metrics.parse(ret)
             if global_step % args.train_log_step == 0:
                 writer.add_scalar(
-                    "batch_loss", ret['loss'], global_step=global_step)
+                    "batch_loss", ret['loss'], global_step)
                 log.info("epoch: %d | step: %d | loss: %.4f " %
                          (e, global_step, ret['loss']))
 
@@ -111,7 +111,7 @@ def train_and_evaluate(exe,
                 for key, value in valid_ret.items():
                     message += "%s %.4f | " % (key, value)
                     writer.add_scalar(
-                        "eval_%s" % key, value, global_step=global_step)
+                        "eval_%s" % key, value, global_step)
                 log.info(message)
 
                 # testing
@@ -120,7 +120,7 @@ def train_and_evaluate(exe,
                 for key, value in test_ret.items():
                     message += "%s %.4f | " % (key, value)
                     writer.add_scalar(
-                        "test_%s" % key, value, global_step=global_step)
+                        "test_%s" % key, value, global_step)
                 log.info(message)
 
         # evaluate after one epoch
@@ -128,7 +128,7 @@ def train_and_evaluate(exe,
         message = "epoch %s valid: " % e
         for key, value in valid_ret.items():
             message += "%s %.4f | " % (key, value)
-            writer.add_scalar("eval_%s" % key, value, global_step=global_step)
+            writer.add_scalar("eval_%s" % key, value, global_step)
         log.info(message)
 
         # testing
@@ -136,7 +136,7 @@ def train_and_evaluate(exe,
         message = "epoch %s test: " % e
         for key, value in test_ret.items():
             message += "%s %.4f | " % (key, value)
-            writer.add_scalar("test_%s" % key, value, global_step=global_step)
+            writer.add_scalar("test_%s" % key, value, global_step)
         log.info(message)
 
         message = "epoch %s best %s result | " % (e, args.eval_metrics)

ogb_examples/linkproppred/ogbl-ppa/monitor/train_monitor.py

@@ -18,7 +18,7 @@ import numpy as np
 import sys
 import os
 import paddle.fluid as F
-from tensorboardX import SummaryWriter
+from pgl.utils.log_writer import LogWriter
 from ogb.linkproppred import Evaluator
 from ogb.linkproppred import LinkPropPredDataset
 
@@ -115,7 +115,7 @@ def train_and_evaluate(exe,
     log_path = os.path.join(output_path, "log")
     _create_if_not_exist(log_path)
 
-    writer = SummaryWriter(log_path)
+    writer = LogWriter(log_path)
 
     best_model = 0
     for e in range(epoch):
@@ -134,7 +134,7 @@ def train_and_evaluate(exe,
             if global_step % train_log_step == 0:
                 for key, value in ret.items():
                     writer.add_scalar(
-                        'train_' + key, value, global_step=global_step)
+                        'train_' + key, value, global_step)
 
             global_step += 1
             if global_step % eval_step == 0:
@@ -149,7 +149,7 @@ def train_and_evaluate(exe,
                 sys.stderr.write(json.dumps(eval_ret, indent=4) + "\n")
 
                 for key, value in eval_ret.items():
-                    writer.add_scalar(key, value, global_step=global_step)
+                    writer.add_scalar(key, value, global_step)
 
                 if eval_ret["valid_hits@100"] > best_model:
                     F.io.save_persistables(
@@ -170,7 +170,7 @@ def train_and_evaluate(exe,
         sys.stderr.write(json.dumps(eval_ret, indent=4) + "\n")
 
         for key, value in eval_ret.items():
-            writer.add_scalar(key, value, global_step=global_step)
+            writer.add_scalar(key, value, global_step)
 
         if eval_ret["valid_hits@100"] > best_model:
             F.io.save_persistables(exe,

pgl/utils/log_writer.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+""" Log writer setup: interface for training visualization.
+"""
+import six
+
+LogWriter = None
+
+if six.PY3:
+    # We highly recommend using VisualDL (https://github.com/PaddlePaddle/VisualDL)
+    # for training visualization in Python 3.
+    from visualdl import LogWriter
+    LogWriter = LogWriter
+elif six.PY2:
+    from tensorboardX import SummaryWriter
+    LogWriter = SummaryWriter
+else:
+    raise ValueError("Not running on Python2 or Python3 ?")

requirements.txt

@@ -4,3 +4,5 @@ cython >= 0.25.2
 #paddlepaddle
 
 redis-py-cluster
+
+visualdl >= 2.0.0b ; python_version >= "3"