deleted prints

parent 6e5f5ee6
......@@ -1130,32 +1130,8 @@ class Engine(EngineBase):
print("learning rate control:", self.learning_rate_control, file=log.v4)
print("pretrain:", self.pretrain, file=log.v4)
#save original data
# #self.train_data = self.train_data.get_data_slice(self.train_data._seq_order, 'target', 0, 10)
# print("print")
# print(self.train_data._seq_order)
# #self.train_data._seq_order = self.train_data._seq_order[:20]
# print(range(self.train_data._data_len))
# print([i for i in range(self.train_data._data_len)])
# #print(self.train_data.compute_seq_difficulty())
# # self.train_data._seq_order = [i for i in range(self.train_data._data_len) if (self.train_data.seq_difficulty[i] < 0.2)]
# self.train_data._seq_order = [i for i in range(150)]
# self.train_data._num_seqs = len(self.train_data._seq_order)
# print("printagain")
# print(self.train_data._seq_order)
# self.train_data.compute_seq_difficulty()
# all_train_data = self.train_data
# inc_by = 0.2
print("In the train")
if self.curriculum_learning['use_curriculum_learning']:
# if 'n_times_per_slice' in self.curriculum_learning:
# #TODO increase learning rate by 3
......@@ -1185,37 +1161,9 @@ class Engine(EngineBase):
if self.train_data.init_seq_order(epoch=self.epoch):
self.dataset_batches.pop("train", None)
if self.curriculum_learning['use_curriculum_learning']:
self.dataset_batches.pop("train", None)
slice_time = time.time()
self.curriculum_learning['use_curriculum_learning'] = self.train_data.make_cur_slice(self.curriculum_learning)
slice_time = time.time() - slice_time
print("make_cur_slice takes %s seconds ---" % (slice_time))
My notes:
-sort the dataset with the script in work/bin
-sort the dataset in returnn (go via _data or slicing in Dataset (parent class) -> has to be a TranslationDataset afterwards
-in the train.o in competence learning is an extraction of the short dataframe with de->en
-den data slice hinkriegen
-schaffen, dass nur ein step pro epoche trainiert wird
-schaffen, dass die auswahl, welche data included wird als vergleich zwischen competence und difficulty werten ist
-the sort methods return from [1,4,7,2] an index list [0,3,1,2] (da wo sie vorher standen)
-ausprobieren, ob die _seq_order reicht um auszuwaehlen, welche data man benutzen will
-newbaseline laeuft
-test mit ersten 100 trainingssasetzen laeuft. sollte overfitten. die logs vergleichen bitte
for dataset_name, dataset in self.get_eval_datasets().items():
Markdown is supported
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment