Make PaddleSlim support PyReader (#19995)

* Make PaddleSlim support PyReader.
* Fix unittest of sensitive pruning.
* Add some assert.

paddle/fluid/API.spec

@@ -492,7 +492,7 @@ paddle.fluid.contrib.QuantizeTranspiler.freeze_program (ArgSpec(args=['self', 'p
 paddle.fluid.contrib.QuantizeTranspiler.training_transpile (ArgSpec(args=['self', 'program', 'startup_program'], varargs=None, keywords=None, defaults=(None, None)), ('document', '6dd9909f10b283ba2892a99058a72884'))
 paddle.fluid.contrib.distributed_batch_reader (ArgSpec(args=['batch_reader'], varargs=None, keywords=None, defaults=None), ('document', 'b60796eb0a481484dd34e345f0eaa4d5'))
 paddle.fluid.contrib.Compressor ('paddle.fluid.contrib.slim.core.compressor.Compressor', ('document', 'a5417774a94aa9ae5560a42b96527e7d'))
-paddle.fluid.contrib.Compressor.__init__ (ArgSpec(args=['self', 'place', 'scope', 'train_program', 'train_reader', 'train_feed_list', 'train_fetch_list', 'eval_program', 'eval_reader', 'eval_feed_list', 'eval_fetch_list', 'eval_func', 'save_eval_model', 'prune_infer_model', 'teacher_programs', 'checkpoint_path', 'train_optimizer', 'distiller_optimizer', 'search_space'], varargs=None, keywords=None, defaults=(None, None, None, None, None, None, None, None, True, None, [], None, None, None, None)), ('document', '05119e0fa0fc07f5cf848ebf0a2cf070'))
+paddle.fluid.contrib.Compressor.__init__ (ArgSpec(args=['self', 'place', 'scope', 'train_program', 'train_reader', 'train_feed_list', 'train_fetch_list', 'eval_program', 'eval_reader', 'eval_feed_list', 'eval_fetch_list', 'eval_func', 'save_eval_model', 'prune_infer_model', 'teacher_programs', 'checkpoint_path', 'train_optimizer', 'distiller_optimizer', 'search_space', 'log_period'], varargs=None, keywords=None, defaults=(None, None, None, None, None, None, None, None, True, None, [], None, None, None, None, 20)), ('document', '26261076fa2140a1367cb4fbf3ac03fa'))
 paddle.fluid.contrib.Compressor.config (ArgSpec(args=['self', 'config_file'], varargs=None, keywords=None, defaults=None), ('document', '780d9c007276ccbb95b292400d7807b0'))
 paddle.fluid.contrib.Compressor.run (ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None), ('document', 'c6e43d6a078d307672283c1f36e04fe9'))
 paddle.fluid.contrib.load_persistables_for_increment (ArgSpec(args=['dirname', 'executor', 'program', 'lookup_table_var', 'lookup_table_var_path'], varargs=None, keywords=None, defaults=None), ('document', '2ab36d4f7a564f5f65e455807ad06c67'))

python/paddle/fluid/contrib/slim/core/compressor.py

@@ -20,6 +20,7 @@ from .... import profiler
 from .... import scope_guard
 from ....data_feeder import DataFeeder
 from ....log_helper import get_logger
+from ....reader import PyReader
 from ..graph import *
 from .config import ConfigFactory
 import numpy as np
@@ -185,10 +186,16 @@ class Context(object):
         s_time = time.time()
         reader = self.eval_reader
         if sampled_rate:
+            assert (not isinstance(reader, Variable))
+            assert (sampled_rate > 0)
+            assert (self.cache_path is not None)
+            _logger.info('sampled_rate: {}; cached_id: {}'.format(sampled_rate,
+                                                                  cached_id))
             reader = cached_reader(reader, sampled_rate, self.cache_path,
                                    cached_id)
 
-        if isinstance(reader, Variable):
+        if isinstance(reader, Variable) or (isinstance(reader, PyReader) and
+                                            (not reader._iterable)):
             reader.start()
             try:
                 while True:
@@ -249,7 +256,8 @@ class Compressor(object):
                  checkpoint_path=None,
                  train_optimizer=None,
                  distiller_optimizer=None,
-                 search_space=None):
+                 search_space=None,
+                 log_period=20):
         """
         Args:
             place(fluid.Place): The device place where the compression job running.
@@ -294,6 +302,7 @@ class Compressor(object):
                                  student-net in fine-tune stage. 
             search_space(slim.nas.SearchSpace): The instance that define the searching space. It must inherite
                               slim.nas.SearchSpace class and overwrite the abstract methods.
+            log_period(int): The period of print log of training.
 
         """
         assert train_feed_list is None or isinstance(
@@ -329,6 +338,8 @@ class Compressor(object):
         self.init_model = None
 
         self.search_space = search_space
+        self.log_period = log_period
+        assert (log_period > 0)
 
     def _add_strategy(self, strategy):
         """
@@ -357,6 +368,7 @@ class Compressor(object):
 
         if 'eval_epoch' in factory.compressor:
             self.eval_epoch = factory.compressor['eval_epoch']
+        assert (self.eval_epoch > 0)
 
     def _init_model(self, context):
         """
@@ -414,7 +426,7 @@ class Compressor(object):
                         else:
                             strategies = pickle.load(
                                 strategy_file, encoding='bytes')
-
+                assert (len(self.strategies) == len(strategies))
                 for s, s1 in zip(self.strategies, strategies):
                     s1.__dict__.update(s.__dict__)
 
@@ -472,7 +484,9 @@ class Compressor(object):
                 context.optimize_graph.program).with_data_parallel(
                     loss_name=context.optimize_graph.out_nodes['loss'])
 
-        if isinstance(context.train_reader, Variable):
+        if isinstance(context.train_reader, Variable) or (
+                isinstance(context.train_reader,
+                           PyReader) and (not context.train_reader._iterable)):
             context.train_reader.start()
             try:
                 while True:
@@ -482,7 +496,7 @@ class Compressor(object):
                     results = executor.run(context.optimize_graph,
                                            context.scope)
                     results = [float(np.mean(result)) for result in results]
-                    if context.batch_id % 20 == 0:
+                    if context.batch_id % self.log_period == 0:
                         _logger.info("epoch:{}; batch_id:{}; {} = {}".format(
                             context.epoch_id, context.batch_id,
                             context.optimize_graph.out_nodes.keys(
@@ -502,7 +516,7 @@ class Compressor(object):
                                        context.scope,
                                        data=data)
                 results = [float(np.mean(result)) for result in results]
-                if context.batch_id % 20 == 0:
+                if context.batch_id % self.log_period == 0:
                     _logger.info("epoch:{}; batch_id:{}; {} = {}".format(
                         context.epoch_id, context.batch_id,
                         context.optimize_graph.out_nodes.keys(

python/paddle/fluid/contrib/slim/prune/prune_strategy.py

@@ -782,7 +782,6 @@ class SensitivePruneStrategy(PruneStrategy):
                 }
 
         metric = None
-
         for param in sensitivities.keys():
             ratio = self.delta_rate
             while ratio < 1:
@@ -825,7 +824,7 @@ class SensitivePruneStrategy(PruneStrategy):
                 # restore pruned parameters
                 for param_name in param_backup.keys():
                     param_t = context.scope.find_var(param_name).get_tensor()
-                    param_t.set(self.param_backup[param_name], context.place)
+                    param_t.set(param_backup[param_name], context.place)
 
 #                pruned_metric = self._eval_graph(context)
 

python/paddle/fluid/contrib/slim/tests/filter_pruning/compress.yaml

@@ -24,11 +24,11 @@ strategies:
         target_ratio: 0.08
         num_steps: 1
         eval_rate: 0.5
-        pruned_params: 'conv6_sep_weights'
+        pruned_params: '_conv6_sep_weights'
         sensitivities_file: 'mobilenet_acc_top1_sensitive.data'
         metric_name: 'acc_top1'
 compressor:
-    epoch: 1
+    epoch: 2
     checkpoint_path: './checkpoints_pruning/'
     strategies:
         - sensitive_pruning_strategy