add vdl to replace tb_paddle (#570)

paddlehub/autofinetune/autoft.py

@@ -22,7 +22,7 @@ import os
 import six
 import time
 
-from tb_paddle import SummaryWriter
+from visualdl import LogWriter
 from paddlehub.common.logger import logger
 from paddlehub.common.utils import mkdir
 from paddlehub.autofinetune.evaluator import REWARD_SUM, TMP_HOME
@@ -67,12 +67,12 @@ class BaseTuningStrategy(object):
             self._output_dir = output_dir
 
         # record the information for the whole auto finetune
-        self.writer = SummaryWriter(logdir=self._output_dir + '/visualization')
+        self.writer = LogWriter(logdir=self._output_dir + '/visualization')
 
         # record the information for per population in all round
         self.writer_pop_trails = []
         for i in range(self.popsize):
-            writer_pop_trail = SummaryWriter(
+            writer_pop_trail = LogWriter(
                 logdir=self._output_dir + '/visualization/pop_{}'.format(i))
             self.writer_pop_trails.append(writer_pop_trail)
 
@@ -285,23 +285,23 @@ class HAZero(BaseTuningStrategy):
         for index, name in enumerate(self.hparams_name_list):
             self.writer.add_scalar(
                 tag="hyperparameter_tuning/" + name,
-                scalar_value=best_hparams[index],
-                global_step=self.round)
+                value=best_hparams[index],
+                step=self.round)
         self.writer.add_scalar(
             tag="hyperparameter_tuning/best_eval_value",
-            scalar_value=self.get_best_eval_value(),
-            global_step=self.round)
+            value=self.get_best_eval_value(),
+            step=self.round)
         for pop_num in range(self.popsize):
             params = self.evaluator.convert_params(params_list[pop_num])
             for index, name in enumerate(self.hparams_name_list):
                 self.writer_pop_trails[pop_num].add_scalar(
                     tag="population_transformation/" + name,
-                    scalar_value=params[index],
-                    global_step=self.round)
+                    value=params[index],
+                    step=self.round)
             self.writer_pop_trails[pop_num].add_scalar(
                 tag="population_transformation/eval_value",
-                scalar_value=(REWARD_SUM - reward_list[pop_num]),
-                global_step=self.round)
+                value=(REWARD_SUM - reward_list[pop_num]),
+                step=self.round)
 
         self.evolution_stratefy.tell(params_list, reward_list)
         self.evolution_stratefy.disp()
@@ -422,23 +422,23 @@ class PSHE2(BaseTuningStrategy):
         for index, name in enumerate(self.hparams_name_list):
             self.writer.add_scalar(
                 tag="hyperparameter_tuning/" + name,
-                scalar_value=best_hparams[index],
-                global_step=self.round)
+                value=best_hparams[index],
+                step=self.round)
         self.writer.add_scalar(
             tag="hyperparameter_tuning/best_eval_value",
-            scalar_value=self.get_best_eval_value(),
-            global_step=self.round)
+            value=self.get_best_eval_value(),
+            step=self.round)
         for pop_num in range(self.popsize):
             params = self.evaluator.convert_params(params_list[pop_num])
             for index, name in enumerate(self.hparams_name_list):
                 self.writer_pop_trails[pop_num].add_scalar(
                     tag="population_transformation/" + name,
-                    scalar_value=params[index],
-                    global_step=self.round)
+                    value=params[index],
+                    step=self.round)
             self.writer_pop_trails[pop_num].add_scalar(
                 tag="population_transformation/eval_value",
-                scalar_value=(REWARD_SUM - reward_list[pop_num]),
-                global_step=self.round)
+                value=(REWARD_SUM - reward_list[pop_num]),
+                step=self.round)
 
         self.estimate_momemtum()
         for i in range(self.popsize):

paddlehub/finetune/task/base_task.py

@@ -32,7 +32,7 @@ else:
 import numpy as np
 import paddle
 import paddle.fluid as fluid
-from tb_paddle import SummaryWriter
+from visualdl import LogWriter
 
 import paddlehub as hub
 from paddlehub.common.paddle_helper import dtype_map, clone_program
@@ -327,7 +327,7 @@ class BaseTask(object):
         self._phases = []
         self._envs = {}
         self._predict_data = None
-        self._tb_writer = None
+        self._vdl_writer = None
 
         # event hooks
         self._hooks = TaskHooks()
@@ -602,16 +602,16 @@ class BaseTask(object):
         return [vars[varname] for varname in self.fetch_list]
 
     @property
-    def tb_writer(self):
+    def vdl_writer(self):
         """
-        get tb_writer for visualization.
+        get vdl_writer for visualization.
         """
         if not os.path.exists(self.config.checkpoint_dir):
             mkdir(self.config.checkpoint_dir)
         tb_log_dir = os.path.join(self.config.checkpoint_dir, "visualization")
-        if not self._tb_writer:
-            self._tb_writer = SummaryWriter(tb_log_dir)
-        return self._tb_writer
+        if not self._vdl_writer:
+            self._vdl_writer = LogWriter(tb_log_dir)
+        return self._vdl_writer
 
     def create_event_function(self, hook_type):
         """
@@ -718,18 +718,19 @@ class BaseTask(object):
         """
         eval_scores, eval_loss, run_speed = self._calculate_metrics(run_states)
         if 'train' in self._envs:
-            self.tb_writer.add_scalar(
+            self.vdl_writer.add_scalar(
                 tag="Loss_{}".format(self.phase),
-                scalar_value=eval_loss,
-                global_step=self._envs['train'].current_step)
+                value=eval_loss,
+                step=self._envs['train'].current_step)
 
         log_scores = ""
         for metric in eval_scores:
             if 'train' in self._envs:
-                self.tb_writer.add_scalar(
+                self.vdl_writer.add_scalar(
                     tag="{}_{}".format(metric, self.phase),
-                    scalar_value=eval_scores[metric],
-                    global_step=self._envs['train'].current_step)
+                    value=eval_scores[metric],
+                    step=self._envs['train'].current_step)
+
             log_scores += "%s=%.5f " % (metric, eval_scores[metric])
         logger.eval(
             "[%s dataset evaluation result] loss=%.5f %s[step/sec: %.2f]" %
@@ -761,16 +762,16 @@ class BaseTask(object):
             run_states (object): the results in train phase
         """
         scores, avg_loss, run_speed = self._calculate_metrics(run_states)
-        self.tb_writer.add_scalar(
+        self.vdl_writer.add_scalar(
             tag="Loss_{}".format(self.phase),
-            scalar_value=avg_loss,
-            global_step=self._envs['train'].current_step)
+            value=avg_loss,
+            step=self._envs['train'].current_step)
         log_scores = ""
         for metric in scores:
-            self.tb_writer.add_scalar(
+            self.vdl_writer.add_scalar(
                 tag="{}_{}".format(metric, self.phase),
-                scalar_value=scores[metric],
-                global_step=self._envs['train'].current_step)
+                value=scores[metric],
+                step=self._envs['train'].current_step)
             log_scores += "%s=%.5f " % (metric, scores[metric])
         logger.train("step %d / %d: loss=%.5f %s[step/sec: %.2f]" %
                      (self.current_step, self.max_train_steps, avg_loss,

requirements.txt

@@ -8,8 +8,7 @@ chardet == 3.0.4
 requests
 flask >= 1.1.0
 flake8
-tb-paddle
-tensorboard >= 1.15
+visualdl == 2.0.0a0
 cma == 2.7.0
 sentencepiece
 nltk