search

kedz / nnsum

An extractive neural network text summarization library for the EMNLP 2018 paper "Content Selection in Deep Learning Models of Summarization" (https://arxiv.org/abs/1810.12343).
108 stars 19 forks source link

ERROR:ignite.engine.engine.Engine:Current run is terminating due to exception: 'NoneType' object has no attribute 'data'. ERROR:ignite.engine.engine.Engine:Engine run is terminating due to exception: 'NoneType' object has no attribute 'data'. #14

Open XuemingQiu opened 4 years ago

XuemingQiu commented 4 years ago

when I try to use this repo to run the AMI dataset , I got this error ,can you help me? My command is: python train_model.py --trainer --train-inputs /home/workspace/IdeaProjects/word2sentExtract/datasets/ami/inputs/train/ --train-labels /home/workspace/IdeaProjects/word2sentExtract/datasets/ami/labels/train/ --valid-inputs /home/workspace/IdeaProjects/word2sentExtract/datasets/ami/inputs/valid/ --valid-labels /home/workspace/IdeaProjects/word2sentExtract/datasets/ami/labels/valid/ --valid-refs /home/workspace/IdeaProjects/word2sentExtract/datasets/ami/human-abstracts/valid/ --model best_model --results results/ --seed 12345678 --emb --enc cnn --ext s2s --bidirectional the error message are follows: `{'train_inputs': PosixPath('/home/workspace/IdeaProjects/word2sentExtract/datasets/ami/inputs/train'), 'train_labels': PosixPath('/home/workspace/IdeaProjects/word2sentExtract/datasets/ami/labels/train'), 'valid_inputs': PosixPath('/home/workspace/IdeaProjects/word2sentExtract/datasets/ami/inputs/valid'), 'valid_labels': PosixPath('/home/workspace/IdeaProjects/word2sentExtract/datasets/ami/labels/valid'), 'valid_refs': PosixPath('/home/workspace/IdeaProjects/word2sentExtract/datasets/ami/human-abstracts/valid'), 'seed': 12345678, 'epochs': 50, 'batch_size': 32, 'gpu': -1, 'teacher_forcing': 25, 'sentence_limit': 50, 'weighted': False, 'loader_workers': 8, 'raml_samples': 25, 'raml_temp': 0.05, 'summary_length': 100, 'remove_stopwords': False, 'shuffle_sents': False, 'model': PosixPath('best_model'), 'results': PosixPath('results')}

{'embedding_size': 200, 'pretrained_embeddings': None, '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: Creating new embeddings with normal initializaion. INFO:root: # Unique Words: 8663 INFO:root: After filtering, # Unique Words: 8665 WARNING:root: Embeddings are randomly initialized but update rule is not 'update-all' INFO:root: EmbeddingContext( (embeddings): Embedding(8665, 200, padding_idx=0) ) Loading training data. Loading validation data. INFO:root: Model parameter initialization started. INFO:root: EmbeddingContext initialization started. INFO:root: Initializing with random normal. INFO:root: EmbeddingContext initialization finished. INFO:root: CNNSentenceEncoder initialization started. INFO:root: filters.0.weight (25,1,1,1,200): Xavier normal init. INFO:root: filters.0.bias (25): constant (0) init. INFO:root: filters.1.weight (25,1,1,2,200): Xavier normal init. INFO:root: filters.1.bias (25): constant (0) init. INFO:root: filters.2.weight (50,1,1,3,200): Xavier normal init. INFO:root: filters.2.bias (50): constant (0) init. INFO:root: filters.3.weight (50,1,1,4,200): Xavier normal init. INFO:root: filters.3.bias (50): constant (0) init. INFO:root: filters.4.weight (50,1,1,5,200): Xavier normal init. INFO:root: filters.4.bias (50): constant (0) init. INFO:root: filters.5.weight (50,1,1,6,200): Xavier normal init. INFO:root: filters.5.bias (50): constant (0) init. INFO:root: CNNSentenceEncoder initialization finished. INFO:root: Seq2SeqSentenceExtractor initialization started. INFO:root: decoder_start (250): random normal init. INFO:root: encoder_rnn.weight_ih_l0 (900,250): Xavier normal init. INFO:root: encoder_rnn.weight_hh_l0 (900,300): Xavier normal init. INFO:root: encoder_rnn.bias_ih_l0 (900): constant (0) init. INFO:root: encoder_rnn.bias_hh_l0 (900): constant (0) init. INFO:root: encoder_rnn.weight_ih_l0_reverse (900,250): Xavier normal init. INFO:root: encoder_rnn.weight_hh_l0_reverse (900,300): Xavier normal init. INFO:root: encoder_rnn.bias_ih_l0_reverse (900): constant (0) init. INFO:root: encoder_rnn.bias_hh_l0_reverse (900): constant (0) init. INFO:root: decoder_rnn.weight_ih_l0 (900,250): Xavier normal init. INFO:root: decoder_rnn.weight_hh_l0 (900,300): Xavier normal init. INFO:root: decoder_rnn.bias_ih_l0 (900): constant (0) init. INFO:root: decoder_rnn.bias_hh_l0 (900): constant (0) init. INFO:root: decoder_rnn.weight_ih_l0_reverse (900,250): Xavier normal init. INFO:root: decoder_rnn.weight_hh_l0_reverse (900,300): Xavier normal init. INFO:root: decoder_rnn.bias_ih_l0_reverse (900): constant (0) init. INFO:root: decoder_rnn.bias_hh_l0_reverse (900): constant (0) init. INFO:root: mlp.0.weight (100,1200): Xavier normal init. INFO:root: mlp.0.bias (100): constant (0) init. INFO:root: mlp.3.weight (1,100): Xavier normal init. INFO:root: mlp.3.bias (1): constant (0) init. INFO:root: Seq2SeqSentenceExtractor initialization finished. INFO:root: Model parameter initialization finished.

INFO:ignite.engine.engine.Engine:Engine run starting with max_epochs=50. ERROR:ignite.engine.engine.Engine:Current run is terminating due to exception: 'NoneType' object has no attribute 'data'. ERROR:ignite.engine.engine.Engine:Engine run is terminating due to exception: 'NoneType' object has no attribute 'data'. Traceback (most recent call last): File "train_model.py", line 79, in main() File "train_model.py", line 76, in main results_path=args["trainer"]["results"]) File "/home/qxm/anaconda3/lib/python3.7/site-packages/nnsum-1.0-py3.7.egg/nnsum/trainer/labels_mle_trainer.py", line 164, in labels_mle_trainer File "/home/qxm/anaconda3/lib/python3.7/site-packages/pytorch_ignite-0.5.0.dev20200721-py3.7.egg/ignite/engine/engine.py", line 658, in run return self._internal_run() File "/home/qxm/anaconda3/lib/python3.7/site-packages/pytorch_ignite-0.5.0.dev20200721-py3.7.egg/ignite/engine/engine.py", line 729, in _internal_run self._handle_exception(e) File "/home/qxm/anaconda3/lib/python3.7/site-packages/pytorch_ignite-0.5.0.dev20200721-py3.7.egg/ignite/engine/engine.py", line 437, in _handle_exception raise e File "/home/qxm/anaconda3/lib/python3.7/site-packages/pytorch_ignite-0.5.0.dev20200721-py3.7.egg/ignite/engine/engine.py", line 697, in _internal_run time_taken = self._run_once_on_dataset() File "/home/qxm/anaconda3/lib/python3.7/site-packages/pytorch_ignite-0.5.0.dev20200721-py3.7.egg/ignite/engine/engine.py", line 795, in _run_once_on_dataset self._handle_exception(e) File "/home/qxm/anaconda3/lib/python3.7/site-packages/pytorch_ignite-0.5.0.dev20200721-py3.7.egg/ignite/engine/engine.py", line 437, in _handle_exception raise e File "/home/qxm/anaconda3/lib/python3.7/site-packages/pytorch_ignite-0.5.0.dev20200721-py3.7.egg/ignite/engine/engine.py", line 778, in _run_once_on_dataset self.state.output = self._process_function(self, self.state.batch) File "/home/qxm/anaconda3/lib/python3.7/site-packages/nnsum-1.0-py3.7.egg/nnsum/trainer/labels_mle_trainer.py", line 188, in _update AttributeError: 'NoneType' object has no attribute 'data' `

How can I fix it? Best wishes for you!

kedz commented 4 years ago

Hi polarsun!

So I think I know what's happening, and it's related to this issue: https://github.com/kedz/nnsum/issues/4#issuecomment-459763674 . When I developed this code, gradients for pytorch tensors were initialized by default. In recent versions of pytorch, tensors do not have a .grad attribute unless they are needed in the backprop computation. Under your settings, you are not using pretrained embeddings but by default the code does not update the word embeddings during training since it assumes that you are using glove or some other pretrained embeddings. Since it is not updating the embeddings during training they do not have a .grad attribute. You can patch the code yourself as described in the link above to fix the issue. Unfortunately I don't have the bandwidth right now to keep this repository up to date with the latest versions of pytorch.

Additionally, since you are not using pretrained embeddings you should probably learn them. To do so, I would add --update-rule update-all after the --emb argument in your command line. I think doing this will actually fix the issue. In any case, while I don't know what your use case is, the ami dataset is quite small and you probably will want to use pretrained embeddings anyway -- see the readme for doing so.

Lemme know if you need more help patching the code!

Best, Chris

On Sun, Jul 26, 2020 at 9:36 AM polarsun notifications@github.com wrote:

when I try to use this repo to run the AMI dataset , I got this error ,can you help me? My command is: python train_model.py --trainer --train-inputs /home/workspace/IdeaProjects/word2sentExtract/datasets/ami/inputs/train/ --train-labels /home/workspace/IdeaProjects/word2sentExtract/datasets/ami/labels/train/ --valid-inputs /home/workspace/IdeaProjects/word2sentExtract/datasets/ami/inputs/valid/ --valid-labels /home/workspace/IdeaProjects/word2sentExtract/datasets/ami/labels/valid/ --valid-refs /home/workspace/IdeaProjects/word2sentExtract/datasets/ami/human-abstracts/valid/ --model best_model --results results/ --seed 12345678 --emb --enc cnn --ext s2s --bidirectional the error message are follows: `{'train_inputs': PosixPath('/home/workspace/IdeaProjects/word2sentExtract/datasets/ami/inputs/train'), 'train_labels': PosixPath('/home/workspace/IdeaProjects/word2sentExtract/datasets/ami/labels/train'), 'valid_inputs': PosixPath('/home/workspace/IdeaProjects/word2sentExtract/datasets/ami/inputs/valid'), 'valid_labels': PosixPath('/home/workspace/IdeaProjects/word2sentExtract/datasets/ami/labels/valid'), 'valid_refs': PosixPath('/home/workspace/IdeaProjects/word2sentExtract/datasets/ami/human-abstracts/valid'), 'seed': 12345678, 'epochs': 50, 'batch_size': 32, 'gpu': -1, 'teacher_forcing': 25, 'sentence_limit': 50, 'weighted': False, 'loader_workers': 8, 'raml_samples': 25, 'raml_temp': 0.05, 'summary_length': 100, 'remove_stopwords': False, 'shuffle_sents': False, 'model': PosixPath('best_model'), 'results': PosixPath('results')}

{'embedding_size': 200, 'pretrained_embeddings': None, '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: Creating new embeddings with normal initializaion. INFO:root: # Unique Words: 8663 INFO:root: After filtering, # Unique Words: 8665 WARNING:root: Embeddings are randomly initialized but update rule is not 'update-all' INFO:root: EmbeddingContext( (embeddings): Embedding(8665, 200, padding_idx=0) ) Loading training data. Loading validation data. INFO:root: Model parameter initialization started. INFO:root: EmbeddingContext initialization started. INFO:root: Initializing with random normal. INFO:root: EmbeddingContext initialization finished. INFO:root: CNNSentenceEncoder initialization started. INFO:root: filters.0.weight (25,1,1,1,200): Xavier normal init. INFO:root: filters.0.bias (25): constant (0) init. INFO:root: filters.1.weight (25,1,1,2,200): Xavier normal init. INFO:root: filters.1.bias (25): constant (0) init. INFO:root: filters.2.weight (50,1,1,3,200): Xavier normal init. INFO:root: filters.2.bias (50): constant (0) init. INFO:root: filters.3.weight (50,1,1,4,200): Xavier normal init. INFO:root: filters.3.bias (50): constant (0) init. INFO:root: filters.4.weight (50,1,1,5,200): Xavier normal init. INFO:root: filters.4.bias (50): constant (0) init. INFO:root: filters.5.weight (50,1,1,6,200): Xavier normal init. INFO:root: filters.5.bias (50): constant (0) init. INFO:root: CNNSentenceEncoder initialization finished. INFO:root: Seq2SeqSentenceExtractor initialization started. INFO:root: decoder_start (250): random normal init. INFO:root: encoder_rnn.weight_ih_l0 (900,250): Xavier normal init. INFO:root: encoder_rnn.weight_hh_l0 (900,300): Xavier normal init. INFO:root: encoder_rnn.bias_ih_l0 (900): constant (0) init. INFO:root: encoder_rnn.bias_hh_l0 (900): constant (0) init. INFO:root: encoder_rnn.weight_ih_l0_reverse (900,250): Xavier normal init. INFO:root: encoder_rnn.weight_hh_l0_reverse (900,300): Xavier normal init. INFO:root: encoder_rnn.bias_ih_l0_reverse (900): constant (0) init. INFO:root: encoder_rnn.bias_hh_l0_reverse (900): constant (0) init. INFO:root: decoder_rnn.weight_ih_l0 (900,250): Xavier normal init. INFO:root: decoder_rnn.weight_hh_l0 (900,300): Xavier normal init. INFO:root: decoder_rnn.bias_ih_l0 (900): constant (0) init. INFO:root: decoder_rnn.bias_hh_l0 (900): constant (0) init. INFO:root: decoder_rnn.weight_ih_l0_reverse (900,250): Xavier normal init. INFO:root: decoder_rnn.weight_hh_l0_reverse (900,300): Xavier normal init. INFO:root: decoder_rnn.bias_ih_l0_reverse (900): constant (0) init. INFO:root: decoder_rnn.bias_hh_l0_reverse (900): constant (0) init. INFO:root: mlp.0.weight (100,1200): Xavier normal init. INFO:root: mlp.0.bias (100): constant (0) init. INFO:root: mlp.3.weight (1,100): Xavier normal init. INFO:root: mlp.3.bias (1): constant (0) init. INFO:root: Seq2SeqSentenceExtractor initialization finished. INFO:root: Model parameter initialization finished.

INFO:ignite.engine.engine.Engine:Engine run starting with max_epochs=50. ERROR:ignite.engine.engine.Engine:Current run is terminating due to exception: 'NoneType' object has no attribute 'data'. ERROR:ignite.engine.engine.Engine:Engine run is terminating due to exception: 'NoneType' object has no attribute 'data'. Traceback (most recent call last): File "train_model.py", line 79, in main() File "train_model.py", line 76, in main results_path=args["trainer"]["results"]) File "/home/qxm/anaconda3/lib/python3.7/site-packages/nnsum-1.0-py3.7.egg/nnsum/trainer/labels_mle_trainer.py", line 164, in labels_mle_trainer File "/home/qxm/anaconda3/lib/python3.7/site-packages/pytorch_ignite-0.5.0.dev20200721-py3.7.egg/ignite/engine/engine.py", line 658, in run return self._internal_run() File "/home/qxm/anaconda3/lib/python3.7/site-packages/pytorch_ignite-0.5.0.dev20200721-py3.7.egg/ignite/engine/engine.py", line 729, in _internal_run self._handle_exception(e) File "/home/qxm/anaconda3/lib/python3.7/site-packages/pytorch_ignite-0.5.0.dev20200721-py3.7.egg/ignite/engine/engine.py", line 437, in _handle_exception raise e File "/home/qxm/anaconda3/lib/python3.7/site-packages/pytorch_ignite-0.5.0.dev20200721-py3.7.egg/ignite/engine/engine.py", line 697, in _internal_run time_taken = self._run_once_on_dataset() File "/home/qxm/anaconda3/lib/python3.7/site-packages/pytorch_ignite-0.5.0.dev20200721-py3.7.egg/ignite/engine/engine.py", line 795, in _run_once_on_dataset self._handle_exception(e) File "/home/qxm/anaconda3/lib/python3.7/site-packages/pytorch_ignite-0.5.0.dev20200721-py3.7.egg/ignite/engine/engine.py", line 437, in _handle_exception raise e File "/home/qxm/anaconda3/lib/python3.7/site-packages/pytorch_ignite-0.5.0.dev20200721-py3.7.egg/ignite/engine/engine.py", line 778, in _run_once_on_dataset self.state.output = self._process_function(self, self.state.batch) File "/home/qxm/anaconda3/lib/python3.7/site-packages/nnsum-1.0-py3.7.egg/nnsum/trainer/labels_mle_trainer.py", line 188, in _update AttributeError: 'NoneType' object has no attribute 'data' `

How can I fix it? Best wishes for you!

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/kedz/nnsum/issues/14, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABAZOM6HD7GA5YSNQBRPJXDR5QWNRANCNFSM4PH7PGPQ .

-- Chris Kedzie PhD Student, Dept. of Computer Science Columbia University email: kedzie@cs.columbia.edu web: www.cs.columbia.edu/~kedzie