fix selfplay data augmentation: add sampled paths

ronghanghu · Sep 17, 2018 · ff43811 · ff43811
1 parent 569b977
commit ff43811
Show file tree

Hide file tree

Showing 3 changed files with 3 additions and 4 deletions.
diff --git a/tasks/R2R/data/download.sh b/tasks/R2R/data/download.sh
@@ -4,3 +4,4 @@ wget https://storage.googleapis.com/bringmeaspoon/R2Rdata/R2R_train.json -P task
 wget https://storage.googleapis.com/bringmeaspoon/R2Rdata/R2R_val_seen.json -P tasks/R2R/data/
 wget https://storage.googleapis.com/bringmeaspoon/R2Rdata/R2R_val_unseen.json -P tasks/R2R/data/
 wget https://storage.googleapis.com/bringmeaspoon/R2Rdata/R2R_test.json -P tasks/R2R/data/
+wget http:https://people.eecs.berkeley.edu/~ronghang/projects/speaker_follower/data_augmentation/R2R_data_augmentation_paths.json -P tasks/R2R/data/
diff --git a/tasks/R2R/selfplay_from_speaker.py b/tasks/R2R/selfplay_from_speaker.py
@@ -90,7 +90,8 @@ def make_arg_parser():
  parser.add_argument("speaker_model_prefix")
  parser.add_argument("pred_results_output_file")
  parser.add_argument("--batch_size", type=int, default=20)
- parser.add_argument("--pred_splits", nargs="+", default=["train_selfplay"])
+ parser.add_argument("--pred_splits", nargs="+",
+ default=["data_augmentation_paths"])
 
  # for rational self-play generation
  parser.add_argument("--follower_model_prefix",

diff --git a/tasks/R2R/utils.py b/tasks/R2R/utils.py
@@ -54,9 +54,6 @@ def distance(pose1, pose2):
 def load_datasets(splits):
  data = []
  for split in splits:
- assert split in ['train', 'train_subset', 'val_seen', 'val_unseen', 'test',
- 'sub_train', 'sub_val_seen', 'sub_val_unseen', 'sub_train_subset',
- 'train_selfplay', 'train_selfplay_small']
  with open('tasks/R2R/data/R2R_%s.json' % split) as f:
  data += json.load(f)
  return data