changed saving method

윤영빈
Commit 31662afe83301c2b7a5ca8756e497b0d558cab36 31662afe 1 parent 756d59d7
Showing 8 changed files with 63 additions and 31 deletions
web/backend/api/views.py
web/backend/yt8m/F/yt8mDataset/savedModel/model_flags.json
web/backend/yt8m/eval.py
web/backend/yt8m/frame_level_models.py
web/backend/yt8m/inference.py
web/backend/yt8m/inference_per_segment.py
web/backend/yt8m/train.py
web/backend/yt8m/utils.py
--- a/web/backend/api/views.py
View file @31662af
+++ b/web/backend/api/views.py
View file @31662af
@@ -93,7 +93,6 @@ class VideoFileUploadView(APIView):
         else:
             return Response(file_serializer.errors, status=status.HTTP_400_BAD_REQUEST)
-    
 class VideoFileList(APIView):
     def get_object(self, pk):
--- a/web/backend/yt8m/F/yt8mDataset/savedModel/model_flags.json 0 → 100644
View file @31662af
+++ b/web/backend/yt8m/F/yt8mDataset/savedModel/model_flags.json 0 → 100644
View file @31662af
+{"model": "NetVLADModelLF", "feature_sizes": "1024,128", "feature_names": "rgb,audio", "frame_features": true, "label_loss": "CrossEntropyLoss"}
\ No newline at end of file
--- a/web/backend/yt8m/eval.py
View file @31662af
+++ b/web/backend/yt8m/eval.py
View file @31662af
@@ -33,17 +33,17 @@ FLAGS = flags.FLAGS
 if __name__ == "__main__":
   # Dataset flags.
   flags.DEFINE_string(
-      "train_dir", "F:/yt8mDataset/savedModel",
+      "train_dir", "F:/yt8mDataset/savedModelaa2",
       "The directory to load the model files from. "
       "The tensorboard metrics files are also saved to this "
       "directory.")
   flags.DEFINE_string(
-      "eval_data_pattern", "",
+      "eval_data_pattern", "F:/yt8mDataset/train2/train*.tfrecord",
       "File glob defining the evaluation dataset in tensorflow.SequenceExample "
       "format. The SequenceExamples are expected to have an 'rgb' byte array "
       "sequence feature as well as a 'labels' int64 context feature.")
   flags.DEFINE_bool(
-      "segment_labels", False,
+      "segment_labels", True,
       "If set, then --eval_data_pattern must be frame-level features (but with"
       " segment_labels). Otherwise, --eval_data_pattern must be aggregated "
       "video-level features. The model must also be set appropriately (i.e. to "
@@ -54,7 +54,7 @@ if __name__ == "__main__":
                        "How many examples to process per batch.")
   flags.DEFINE_integer("num_readers", 8,
                        "How many threads to use for reading input files.")
-  flags.DEFINE_boolean("run_once", False, "Whether to run eval only once.")
+  flags.DEFINE_boolean("run_once", True, "Whether to run eval only once.")
   flags.DEFINE_integer("top_k", 5, "How many predictions to output per video.")
--- a/web/backend/yt8m/frame_level_models.py
View file @31662af
+++ b/web/backend/yt8m/frame_level_models.py
View file @31662af
@@ -46,7 +46,7 @@ flags.DEFINE_string(
     "Some Frame-Level models can be decomposed into a "
     "generalized pooling operation followed by a "
     "classifier layer")
-flags.DEFINE_integer("lstm_cells", 512, "Number of LSTM cells.")
+flags.DEFINE_integer("lstm_cells", 1024, "Number of LSTM cells.")
 flags.DEFINE_integer("lstm_layers", 2, "Number of LSTM layers.")
--- a/web/backend/yt8m/inference.py
View file @31662af
+++ b/web/backend/yt8m/inference.py
View file @31662af
@@ -23,7 +23,7 @@ import tarfile
 import tempfile
 import time
 import numpy as np
-
+import ssl
 import readers
 from six.moves import urllib
 import tensorflow as tf
@@ -39,11 +39,11 @@ FLAGS = flags.FLAGS
 if __name__ == "__main__":
   # Input
   flags.DEFINE_string(
-      "train_dir", "", "The directory to load the model files from. We assume "
+      "train_dir", "/mnt/f/yt8mDataset/savedModelaa", "The directory to load the model files from. We assume "
       "that you have already run eval.py onto this, such that "
       "inference_model.* files already exist.")
   flags.DEFINE_string(
-      "input_data_pattern", "",
+      "input_data_pattern", "/mnt/f/yt8mDataset/train2/train*.tfrecord",
       "File glob defining the evaluation dataset in tensorflow.SequenceExample "
       "format. The SequenceExamples are expected to have an 'rgb' byte array "
       "sequence feature as well as a 'labels' int64 context feature.")
@@ -60,7 +60,7 @@ if __name__ == "__main__":
       "be created and the contents of the .tgz file will be "
       "untarred here.")
   flags.DEFINE_bool(
-      "segment_labels", False,
+      "segment_labels", True,
       "If set, then --input_data_pattern must be frame-level features (but with"
       " segment_labels). Otherwise, --input_data_pattern must be aggregated "
       "video-level features. The model must also be set appropriately (i.e. to "
@@ -69,18 +69,18 @@ if __name__ == "__main__":
                        "Limit total number of segment outputs per entity.")
   flags.DEFINE_string(
       "segment_label_ids_file",
-      "https://raw.githubusercontent.com/google/youtube-8m/master/segment_label_ids.csv",
+      "/mnt/e/khuhub/2015104192/web/backend/yt8m/vocabulary.csv",
       "The file that contains the segment label ids.")
   # Output
-  flags.DEFINE_string("output_file", "", "The file to save the predictions to.")
+  flags.DEFINE_string("output_file", "/mnt/f/yt8mDataset/kaggle_solution_validation_temp.csv", "The file to save the predictions to.")
   flags.DEFINE_string(
       "output_model_tgz", "",
       "If given, should be a filename with a .tgz extension, "
       "the model graph and checkpoint will be bundled in this "
       "gzip tar. This file can be uploaded to Kaggle for the "
       "top 10 participants.")
-  flags.DEFINE_integer("top_k", 20, "How many predictions to output per video.")
+  flags.DEFINE_integer("top_k", 5, "How many predictions to output per video.")
   # Other flags.
   flags.DEFINE_integer("batch_size", 512,
@@ -108,18 +108,15 @@ def format_lines(video_ids, predictions, top_k, whitelisted_cls_mask=None):
 def get_input_data_tensors(reader, data_pattern, batch_size, num_readers=1):
   """Creates the section of the graph which reads the input data.
-
   Args:
     reader: A class which parses the input data.
     data_pattern: A 'glob' style path to the data files.
     batch_size: How many examples to process at a time.
     num_readers: How many I/O threads to use.
-
   Returns:
     A tuple containing the features tensor, labels tensor, and optionally a
     tensor containing the number of frames per video. The exact dimensions
     depend on the reader being used.
-
   Raises:
     IOError: If no files matching the given pattern were found.
   """
@@ -243,6 +240,12 @@ def inference(reader, train_dir, data_pattern, out_file_location, batch_size,
         whitelisted_cls_mask = np.zeros((predictions_tensor.get_shape()[-1],),
                                         dtype=np.float32)
         segment_label_ids_file = FLAGS.segment_label_ids_file
+        """
+        if segment_label_ids_file.startswith("http"):
+          logging.info("Retrieving segment ID whitelist files from %s...",
+                       segment_label_ids_file)
+          segment_label_ids_file, _ = urllib.request.urlretrieve(segment_label_ids_file)
+        """
         if segment_label_ids_file.startswith("http"):
           logging.info("Retrieving segment ID whitelist files from %s...",
                        segment_label_ids_file)
@@ -307,6 +310,7 @@ def inference(reader, train_dir, data_pattern, out_file_location, batch_size,
         heaps = {}
         out_file.seek(0, 0)
         for line in out_file:
+          print(line)
           segment_id, preds = line.decode("utf8").split(",")
           if segment_id == "VideoId":
             # Skip the headline.
--- a/web/backend/yt8m/inference_per_segment.py
View file @31662af
+++ b/web/backend/yt8m/inference_per_segment.py
View file @31662af
@@ -41,11 +41,11 @@ FLAGS = flags.FLAGS
 if __name__ == "__main__":
   # Input
   flags.DEFINE_string(
-      "train_dir", "", "The directory to load the model files from. We assume "
+      "train_dir", "E:/savedModel", "The directory to load the model files from. We assume "
       "that you have already run eval.py onto this, such that "
       "inference_model.* files already exist.")
   flags.DEFINE_string(
-      "input_data_pattern", "",
+      "input_data_pattern", "F:/yt8mDataset/test/train*.tfrecord",
       "File glob defining the evaluation dataset in tensorflow.SequenceExample "
       "format. The SequenceExamples are expected to have an 'rgb' byte array "
       "sequence feature as well as a 'labels' int64 context feature.")
@@ -62,7 +62,7 @@ if __name__ == "__main__":
       "be created and the contents of the .tgz file will be "
       "untarred here.")
   flags.DEFINE_bool(
-      "segment_labels", False,
+      "segment_labels", True,
       "If set, then --input_data_pattern must be frame-level features (but with"
       " segment_labels). Otherwise, --input_data_pattern must be aggregated "
       "video-level features. The model must also be set appropriately (i.e. to "
@@ -75,7 +75,7 @@ if __name__ == "__main__":
       "The file that contains the segment label ids.")
   # Output
-  flags.DEFINE_string("output_file", "", "The file to save the predictions to.")
+  flags.DEFINE_string("output_file", "F:/yt8mDataset/kaggle_solution_validation_temp.csv", "The file to save the predictions to.")
   flags.DEFINE_string(
       "output_model_tgz", "",
       "If given, should be a filename with a .tgz extension, "
@@ -85,7 +85,7 @@ if __name__ == "__main__":
   flags.DEFINE_integer("top_k", 5, "How many predictions to output per video.")
   # Other flags.
-  flags.DEFINE_integer("batch_size", 32,
+  flags.DEFINE_integer("batch_size", 16,
                        "How many examples to process per batch.")
   flags.DEFINE_integer("num_readers", 1,
                        "How many threads to use for reading input files.")
@@ -269,6 +269,7 @@ def inference(reader, train_dir, data_pattern, out_file_location, batch_size,
             except ValueError:
               # Simply skip the non-integer line.
               continue
+          fobj.close()
     out_file.write(u"VideoId,LabelConfidencePairs\n".encode("utf8"))
@@ -286,8 +287,10 @@ def inference(reader, train_dir, data_pattern, out_file_location, batch_size,
     vocabs.close()
     try:
       while not coord.should_stop():
+        print("CAME IN")
         video_id_batch_val, video_batch_val, num_frames_batch_val = sess.run(
             [video_id_batch, video_batch, num_frames_batch])
+        print("SESS OUT")
         if FLAGS.segment_labels:
           results = get_segments(video_batch_val, num_frames_batch_val, 5)
           video_segment_ids = results["video_segment_ids"]
@@ -333,6 +336,7 @@ def inference(reader, train_dir, data_pattern, out_file_location, batch_size,
         segment_classes = []
         cls_result_arr = []
         cls_score_dict = {}
+        resultList = []
         out_file.seek(0, 0)
         old_seg_name = '0000'
         counter = 0
@@ -344,12 +348,30 @@ def inference(reader, train_dir, data_pattern, out_file_location, batch_size,
           if segment_id == "VideoId":
             # Skip the headline.
             continue
-
+          print(line)
           preds = preds.split(" ")
+          """
           pred_cls_ids = [int(preds[idx]) for idx in range(0, len(preds), 2)]
           pred_cls_scores = [float(preds[idx]) for idx in range(1, len(preds), 2)]
+          """
+          """
+          denom = 1.0
+          for i in range(0,top_k):
+            denom = denom + float(preds[(i*2) + 1])
+          print("DENOM = ",denom)
+          for i in range(0,top_k):
+            preds[(i*2) + 1] = float(preds[(i*2) + 1])/denom
+          """
+          
+          segment_id = "{0}_{1}".format(str(segment_id.split(":")[0]),str(int(segment_id.split(":")[1])/5))
+          resultList.append("{0},{1},{2},{3},{4},{5}".format(segment_id,
+                                                             preds[0]+","+str(preds[1]),
+                                                             preds[2]+","+str(preds[3]),
+                                                             preds[4]+","+str(preds[5]),
+                                                             preds[6]+","+str(preds[7]),
+                                                             preds[8]+","+str(preds[9])))
           #=======================================
-          segment_id = str(segment_id.split(":")[0])
+          """
           if segment_id not in segment_id_list:
             segment_id_list.append(str(segment_id))
             segment_classes.append("")
@@ -372,12 +394,12 @@ def inference(reader, train_dir, data_pattern, out_file_location, batch_size,
           cls_arr = item.split(" ")[:-1]
           cls_arr = list(map(int,cls_arr))      
-          cls_arr = sorted(cls_arr) #클래스별로 정렬
+          cls_arr = sorted(cls_arr) #sort by class
           result_string = ""
           temp = cls_score_dict[segs]
-          temp= sorted(temp.items(), key=operator.itemgetter(1), reverse=True) #밸류값 기준으로 정렬
+          temp= sorted(temp.items(), key=operator.itemgetter(1), reverse=True) #sort by value
           demoninator = float(temp[0][1] + temp[1][1] + temp[2][1] + temp[3][1] + temp[4][1])
           #for item in temp: 
           for itemIndex in range(0, top_k):
@@ -387,11 +409,16 @@ def inference(reader, train_dir, data_pattern, out_file_location, batch_size,
             result_string = result_string + normalized_tag + ":" + format(temp[itemIndex][1]/demoninator,".3f") + ","
           cls_result_arr.append(result_string[:-1])
-          logging.info(segs + " : " + result_string[:-1])
+          #logging.info(segs + " : " + result_string[:-1])
+        
         #=======================================
         final_out_file.write("vid_id,segment1,segment2,segment3,segment4,segment5\n")
         for seg_id, class_indcies in zip(segment_id_list, cls_result_arr):
           final_out_file.write("%s,%s\n" %(seg_id, str(class_indcies)))
+        """
+        final_out_file.write("vid_id,segment1,pred1,segment2,pred2,segment3,pred3,segment4,pred4,segment5,pred5\n")
+        for resultInfo in resultList:
+              final_out_file.write(resultInfo)
         final_out_file.close()
       out_file.close()
@@ -410,7 +437,7 @@ def main(unused_argv):
       os.makedirs(FLAGS.untar_model_dir)
     tarfile.open(FLAGS.input_model_tgz).extractall(FLAGS.untar_model_dir)
     FLAGS.train_dir = FLAGS.untar_model_dir
-
+  print("TRAIN DIR ", FLAGS.input_data_pattern)
   flags_dict_file = os.path.join(FLAGS.train_dir, "model_flags.json")
   if not file_io.file_exists(flags_dict_file):
     raise IOError("Cannot find %s. Did you run eval.py?" % flags_dict_file)
--- a/web/backend/yt8m/train.py
View file @31662af
+++ b/web/backend/yt8m/train.py
View file @31662af
@@ -75,7 +75,7 @@ if __name__ == "__main__":
   flags.DEFINE_integer(
       "num_gpu", 1, "The maximum number of GPU devices to use for training. "
       "Flag only applies if GPUs are installed")
-  flags.DEFINE_integer("batch_size", 64,
+  flags.DEFINE_integer("batch_size", 16,
                        "How many examples to process per batch for training.")
   flags.DEFINE_string("label_loss", "CrossEntropyLoss",
                       "Which loss function to use for training the model.")
@@ -94,13 +94,13 @@ if __name__ == "__main__":
       "Multiply current learning rate by learning_rate_decay "
       "every learning_rate_decay_examples.")
   flags.DEFINE_integer(
-      "num_epochs", 5, "How many passes to make over the dataset before "
+      "num_epochs", 100, "How many passes to make over the dataset before "
       "halting training.")
   flags.DEFINE_integer(
-      "max_steps", None,
+      "max_steps", 100,
       "The maximum number of iterations of the training loop.")
   flags.DEFINE_integer(
-      "export_model_steps", 5,
+      "export_model_steps", 10,
       "The period, in number of steps, with which the model "
       "is exported for batch prediction.")
--- a/web/backend/yt8m/utils.py
View file @31662af
+++ b/web/backend/yt8m/utils.py
View file @31662af
@@ -154,6 +154,7 @@ def GetListOfFeatureNamesAndSizes(feature_names, feature_sizes):
     List of the feature names and list of the dimensionality of each feature.
     Elements in the first/second list are strings/integers.
   """
+  feature_sizes = str(feature_sizes)
   list_of_feature_names = [
       feature_names.strip() for feature_names in feature_names.split(",")
   ]