#257 done.

arekit/common/experiment/labeling.py

@@ -17,10 +17,6 @@ def __init__(self, uint_labeled_ids):
         self.__original_uint_labels = collections.OrderedDict(uint_labeled_ids)
         self.__assigned_uint_labels = {}
 
-    # TODO. #257. Remove.
-    def is_empty(self):
-        return len(self.__assigned_uint_labels) == 0
-
     def assign_uint_label(self, uint_label, sample_row_id):
         """
         Optionally applies the label.
@@ -32,10 +28,6 @@ def assign_uint_label(self, uint_label, sample_row_id):
         if sample_row_id not in self.__assigned_uint_labels:
             self.__assigned_uint_labels[sample_row_id] = uint_label
 
-    # TODO. #257. Remove.
-    def reset_labels(self):
-        self.__assigned_uint_labels.clear()
-
     def iter_non_duplicated_labeled_sample_row_ids(self):
         for sample_id, _ in self.__original_uint_labels.items():
             yield sample_id, self.__assigned_uint_labels[sample_id]

arekit/contrib/networks/core/ctx_inference.py

@@ -15,10 +15,10 @@
 
 class InferenceContext(object):
 
-    def __init__(self, labeled_samples_dict, bags_collections_dict):
-        assert(isinstance(labeled_samples_dict, dict))
+    def __init__(self, sample_label_pairs_dict, bags_collections_dict):
+        assert(isinstance(sample_label_pairs_dict, dict))
         assert(isinstance(bags_collections_dict, dict))
-        self.__labeled_samples_dict = labeled_samples_dict
+        self.__sample_label_pairs_dict = sample_label_pairs_dict
         self.__bags_collections_dict = bags_collections_dict
         self.__train_stat_uint_labeled_sample_row_ids = None
 
@@ -29,8 +29,8 @@ def BagsCollections(self):
         return self.__bags_collections_dict
 
     @property
-    def LabeledSamplesCollections(self):
-        return self.__labeled_samples_dict
+    def SampleIdAndLabelPairs(self):
+        return self.__sample_label_pairs_dict
 
     @property
     def HasNormalizedWeights(self):
@@ -40,7 +40,7 @@ def HasNormalizedWeights(self):
 
     @classmethod
     def create_empty(cls):
-        return cls(labeled_samples_dict={}, bags_collections_dict={})
+        return cls(sample_label_pairs_dict={}, bags_collections_dict={})
 
     def initialize(self, dtypes, create_samples_view_func, has_model_predefined_state,
                    vocab, labels_count, bags_collection_type, bag_size, input_shapes):
@@ -71,8 +71,7 @@ def initialize(self, dtypes, create_samples_view_func, has_model_predefined_stat
 
             # Saving into dictionaries.
             self.__bags_collections_dict[data_type] = bags_collection
-            self.__labeled_samples_dict[data_type] = LabeledCollection(
-                uint_labeled_ids=uint_labeled_sample_row_ids)
+            self.__sample_label_pairs_dict[data_type] = list(uint_labeled_sample_row_ids)
 
             if data_type == DataType.Train:
                 self.__train_stat_uint_labeled_sample_row_ids = uint_labeled_sample_row_ids

arekit/contrib/networks/core/model_ctx.py

@@ -56,8 +56,8 @@ def __set_optimiser_value(self, value):
     def get_bags_collection(self, data_type):
         return self.__inference_ctx.BagsCollections[data_type]
 
-    def get_labeled_samples_collection(self, data_type):
-        return self.__inference_ctx.LabeledSamplesCollections[data_type]
+    def get_sample_id_label_pairs(self, data_type):
+        return self.__inference_ctx.SampleIdAndLabelPairs[data_type]
 
     def get_hidden_parameters(self):
         names = []

arekit/contrib/networks/core/pipeline_predict.py

@@ -3,29 +3,6 @@
 
 
 class EpochLabelsPredictorPipelineItem(EpochHandlingPipelineItem):
-    """ Considering to treat feed dictionary.
-    """
-
-    def __init__(self):
-        super(EpochLabelsPredictorPipelineItem, self).__init__()
-        self.__labeled_samples = None
-
-    @property
-    def LabeledSamples(self):
-        return self.__labeled_samples
-
-    def before_epoch(self, model_context, data_type):
-        super(EpochLabelsPredictorPipelineItem, self).before_epoch(model_context=model_context,
-                                                                   data_type=data_type)
-
-        # Select the appropriate labels collection.
-        self.__labeled_samples = self._context.InferenceContext.LabeledSamplesCollections[data_type]
-
-        # Clear and assert the correctness.
-        # TODO. #257. Remove. Create new instance instead.
-        self.__labeled_samples.reset_labels()
-        # TODO. #257. Remove. Create new instance instead.
-        assert(self.__labeled_samples.is_empty())
 
     def apply(self, pipeline_ctx):
         assert(isinstance(pipeline_ctx, PipelineContext))
@@ -34,8 +11,4 @@ def apply(self, pipeline_ctx):
         feed_dict = self._context.create_feed_dict(minibatch=minibatch, data_type=self._data_type)
         uint_labels = self._context.Session.run(self._context.Network.Labels, feed_dict=feed_dict)
 
-        # Apply labeling.
-        for bag_index, bag in enumerate(minibatch.iter_by_bags()):
-            uint_label = int(uint_labels[bag_index])
-            for sample in bag:
-                self.__labeled_samples.assign_uint_label(uint_label, sample.ID)
+        pipeline_ctx.update("uint_labels", uint_labels)

arekit/contrib/networks/core/pipeline_predict_labeling.py

@@ -0,0 +1,33 @@
+from arekit.common.experiment.labeling import LabeledCollection
+from arekit.common.pipeline.context import PipelineContext
+from arekit.contrib.networks.core.pipeline_epoch import EpochHandlingPipelineItem
+
+
+class EpochLabelsCollectorPipelineItem(EpochHandlingPipelineItem):
+
+    def __init__(self):
+        super(EpochLabelsCollectorPipelineItem, self).__init__()
+        self.__labeled_samples = None
+
+    @property
+    def LabeledSamples(self):
+        return self.__labeled_samples
+
+    def before_epoch(self, model_context, data_type):
+        super(EpochLabelsCollectorPipelineItem, self).before_epoch(model_context=model_context,
+                                                                   data_type=data_type)
+        pairs = self._context.InferenceContext.SampleIdAndLabelPairs[data_type]
+        self.__labeled_samples = LabeledCollection(uint_labeled_ids=pairs)
+
+    def apply(self, pipeline_ctx):
+        assert(isinstance(pipeline_ctx, PipelineContext))
+        assert("uint_labels" in pipeline_ctx)
+
+        minibatch = pipeline_ctx.provide("src")
+        uint_labels = pipeline_ctx.provide("uint_labels")
+
+        # Apply labeling.
+        for bag_index, bag in enumerate(minibatch.iter_by_bags()):
+            uint_label = int(uint_labels[bag_index])
+            for sample in bag:
+                self.__labeled_samples.assign_uint_label(uint_label, sample.ID)

arekit/contrib/networks/run_training.py

@@ -12,6 +12,7 @@
 from arekit.contrib.networks.core.pipeline_fit import MinibatchFittingPipelineItem
 from arekit.contrib.networks.core.pipeline_keep_hidden import MinibatchHiddenFetcherPipelineItem
 from arekit.contrib.networks.core.pipeline_predict import EpochLabelsPredictorPipelineItem
+from arekit.contrib.networks.core.pipeline_predict_labeling import EpochLabelsCollectorPipelineItem
 from arekit.contrib.networks.shapes import NetworkInputShapes
 from arekit.contrib.networks.utils import rm_dir_contents
 
@@ -103,6 +104,7 @@ def _handle_iteration(self, it_index):
             callback=callback,
             predict_pipeline=[
                 EpochLabelsPredictorPipelineItem(),
+                EpochLabelsCollectorPipelineItem(),
                 MinibatchHiddenFetcherPipelineItem()
             ],
             fit_pipeline=[MinibatchFittingPipelineItem()])

examples/run_text_infer.py

@@ -14,6 +14,7 @@
 from arekit.contrib.networks.core.pipeline_fit import MinibatchFittingPipelineItem
 from arekit.contrib.networks.core.pipeline_keep_hidden import MinibatchHiddenFetcherPipelineItem
 from arekit.contrib.networks.core.pipeline_predict import EpochLabelsPredictorPipelineItem
+from arekit.contrib.networks.core.pipeline_predict_labeling import EpochLabelsCollectorPipelineItem
 from arekit.contrib.networks.core.predict.provider import BasePredictProvider
 from arekit.contrib.networks.core.predict.tsv_writer import TsvPredictWriter
 from arekit.contrib.networks.factory import create_network_and_network_config_funcs
@@ -153,14 +154,15 @@
         callback=NetworkCallback(),
         predict_pipeline=[
             EpochLabelsPredictorPipelineItem(),
+            EpochLabelsCollectorPipelineItem(),
             MinibatchHiddenFetcherPipelineItem()
         ],
         fit_pipeline=[MinibatchFittingPipelineItem()])
 
     model.predict(do_compile=True)
 
     # Gather annotated contexts onto document level.
-    item = model.from_predicted(EpochLabelsPredictorPipelineItem)
+    item = model.from_predicted(EpochLabelsCollectorPipelineItem)
     labeled_samples = item.LabeledSamples
 
     predict_provider = BasePredictProvider()