somethings goes wrong after loading data about nnsum

[avineshpvs]

Since the very recent version pytorch, the gradient buffers are initialized only when needed.
I could find a workaround by adding a check before clipping the gradients in nnsum/trainer/labels_mle_trainer.py

if hasattr(param.grad, 'data'):

nnsum/nnsum/trainer/labels_mle_trainer.py

Lines 187 to 188 in 4660dc2

	for param in model.parameters():
	param.grad.data.clamp_(-grad_clip, grad_clip)

[tlifcen]

Hello,
I'd like to test the cnn-dailymail dataset, but I can not download the dataset. So, I used reddit dataset to test the code. I tried with the command:

python script_bin/train_model.py \
    --trainer --train-inputs  ../dataset/summarization/nnsum/reddit/reddit/inputs/train\
              --train-labels ../dataset/summarization/nnsum/reddit/reddit/abs_lables/train \
              --valid-inputs ../dataset/summarization/nnsum/reddit/reddit/inputs/valid \
              --valid-labels ../dataset/summarization/nnsum/reddit/reddit/abs_lables/valid \
              --valid-refs ../dataset/summarization/nnsum/reddit/reddit/human-abstracts/valid \
              --weighted \
              --gpu 0 \
              --model ../dataset/summarization/nnsum/reddit/model \
              --results ../dataset/summarization/nnsum/reddit/val_score \
              --seed 12345678 \
    --emb --pretrained-embeddings ../dataset/embedding/eng_word_embedding.word2vec.vec \
    --enc cnn \
    --ext s2s --bidirectional

However, some reference problems occurred:

{'train_inputs': PosixPath('../dataset/summarization/nnsum/reddit/reddit/inputs/train'), 'train_labels': PosixPath('../dataset/summarization/nnsum/reddit/reddit/abs_lables/train'), 'valid_inputs': PosixPath('../dataset/summarization/nnsum/reddit/reddit/inputs/valid'), 'valid_labels': PosixPath('../dataset/summarization/nnsum/reddit/reddit/abs_lables/valid'), 'valid_refs': PosixPath('../dataset/summarization/nnsum/reddit/reddit/human-abstracts/valid'), 'seed': 12345678, 'epochs': 50, 'batch_size': 32, 'gpu': 0, 'teacher_forcing': 25, 'sentence_limit': 50, 'weighted': True, 'loader_workers': 8, 'raml_samples': 25, 'raml_temp': 0.05, 'summary_length': 100, 'remove_stopwords': False, 'shuffle_sents': False, 'model': PosixPath('../dataset/summarization/nnsum/reddit/model'), 'results': PosixPath('../dataset/summarization/nnsum/reddit/val_score')}

{'embedding_size': 200, 'pretrained_embeddings': '../dataset/embedding/eng_word_embedding.word2vec.vec', 'top_k': None, 'at_least': 1, 'word_dropout': 0.0, 'embedding_dropout': 0.25, 'update_rule': 'fix-all', 'filter_pretrained': False}

{'dropout': 0.25, 'filter_windows': [1, 2, 3, 4, 5, 6], 'feature_maps': [25, 25, 50, 50, 50, 50], 'OPT': 'cnn'}

{'hidden_size': 300, 'bidirectional': True, 'rnn_dropout': 0.25, 'num_layers': 1, 'cell': 'gru', 'mlp_layers': [100], 'mlp_dropouts': [0.25], 'OPT': 's2s'}
Initializing vocabulary and embeddings.
INFO:root: Reading pretrained embeddings from ../dataset/embedding/eng_word_embedding.word2vec.vec
INFO:root: Read 559185 embeddings of size 200
INFO:root: EmbeddingContext(
  (embeddings): Embedding(559185, 200, padding_idx=0)
)
Loading training data.
Loading validation data.
Traceback (most recent call last):
  File "script_bin/train_model.py", line 79, in <module>
    main()
  File "script_bin/train_model.py", line 48, in main
    sentence_limit=args["trainer"]["sentence_limit"])
  File "/home/constant/anaconda3/lib/python3.7/site-packages/nnsum-1.0-py3.7.egg/nnsum/data/summarization_dataset.py", line 30, in __init__
  File "/home/constant/anaconda3/lib/python3.7/site-packages/nnsum-1.0-py3.7.egg/nnsum/data/summarization_dataset.py", line 59, in _collect_references
Exception: No references found for example id: 12he9h.32

I looked the data files, there are 12he9h.32.a.txt and 12he9h.32.d.txt in the path reddit/human-abstracts/valid. I do not know where the problem is.

Thanks for any help! @kedz

[alexookah]

@tlifcen i encountered the same problem.
when you run python script_bin/train_model.py make sure you are using the correct paths.

assume your current directory is from python_script directory.