2. Recursion in Recursion: Two-Level Nested Recursion for Length Generalization with Scalability - Jishnu Ray Chowdhury, Cornelia Caragea NeurIPS 2023
- CRvNN-related models are adapted from: https://github.com/JRC1995/Continuous-RvNN
models/encoder/OrderedMemory.pyadapted from: https://github.com/yikangshen/Ordered-Memory- BT-GRC, and GT-GRC related models are adapted from: https://github.com/jihunchoi/unsupervised-treelstm
optimizers/ranger.pyadapted from: https://github.com/anoidgit/transformer/blob/master/optm/ranger.pyextensions/, models/modules/S4.pyis copied from: https://github.com/HazyResearch/state-spaces/tree/02e1ba8731ceea90db8343d6e9e150509607bec7preprocess/IMBD_lra.pyis adapted from https://github.com/google-research/long-range-arenamodels/encoders/MEGA.pyandfairseq/adapted from https://github.com/facebookresearch/mega
- pytorch 1.10.0
- pytorch-lightning 1.9.3
- tqdm 4.62.3
- tensorflow-datasets 4.5.2
- typing_extensions 4.5.0
- pykeops 2.1.1
- jsonlines 2.0.0
- einops 0.6.0
- torchtext 0.8.1
- Put the Logical Inference data files (train0,train1,train2,.....test12) (downloaded from https://github.com/yikangshen/Ordered-Memory/tree/master/data/propositionallogic) in data/proplogic/
- Download the ListOps data (along with extrapolation data) from the urls here: https://github.com/facebookresearch/latent-treelstm/blob/master/data/listops/external/urls.txt and put the tsv files in data/listops/
- Run all the make*.py files in data/listops/ to create relevant splits (exact splits used in the paper will be released later)
- Download LRA (https://github.com/google-research/long-range-arena) dataset
- From LRA dataset put the ListOps basic_test.tsv (LRA test set) in data/listops
- From LRA dataset put the ListOps basic_train.tsv, basic_val.tsv, and basic_test.tsv in data/listops_lra
- From LRA dataset's Retrieval task put new_aan_pairs.train.tsv, new_aan_pairs.eval.tsv, and new_aan_pairs.test.tsv in data/AAN.
- Download IMDB original split from here: http://ai.stanford.edu/~amaas/data/sentiment/aclImdb_v1.tar.gz, extract and put the acllmdb folder in data/IMDB/
- Download IMDB contrast set from here: https://github.com/allenai/contrast-sets/tree/main/IMDb/data, put the dev_original.tsv, dev_contrast.tsv and test_contrast.tsv in data/IMDB/
- Download IMDB counterfactual test set from here: https://github.com/acmi-lab/counterfactually-augmented-data/blob/master/sentiment/new/test.tsv; rename it to "test_counterfactual.tsv". Put it in data/IMDB
- Download MNLI datasets (https://cims.nyu.edu/~sbowman/multinli/) and put them in data/MNLI/
- Download MNLI stress tests (https://abhilasharavichander.github.io/NLI_StressTest/) and put them in data/MNLI/
- Download Conjunctive NLI dev set from here (https://github.com/swarnaHub/ConjNLI/tree/master/data/NLI) and put it (conj_dev.tsv) in data/MNLI/
- Put the SNLI files (downloaded and extracted from here: https://nlp.stanford.edu/projects/snli/) in data/SNLI/
- Download https://nlp.stanford.edu/projects/snli/snli_1.0_test_hard.jsonl and put it in data/SNLI/
- Extract dataset.jsonl from here: https://github.com/BIU-NLP/Breaking_NLI and put it in data/SNLI/
- Download the test.tsv from here: https://github.com/acmi-lab/counterfactually-augmented-data/tree/master/NLI/revised_combined and put it in data/SNLI/revised_combined/
- Download QQP (quora_duplicate_questions.tsv) from here: https://quoradata.quora.com/First-Quora-Dataset-Release-Question-Pairs, and put it in data/QQP/
- Follow instructions from here (https://github.com/google-research-datasets/paws) to generate PAWS QQP dev_and_test.tsv and put it in data/QQP/PAWS_QQP/
- Follow instructions from here (https://github.com/google-research-datasets/paws) to generate PAWS WIKI test.tsv and put it in data/QQP/PAWS_WIKI/
- Put glove.840B.300d.txt (download glove.840B.300d.zip from here:https://nlp.stanford.edu/projects/glove/) in embeddings/glove/
- Go to preprocess/ and run each preprocess files to preprocess the corresponding data (process_SNLI_addon.py must be run after process_SNLI.py; otherwise no order requirement)
You can verify if the data is properly set up from the tree below.
├───data
│ ├───AAN
│ │ new_aan_pairs.eval.tsv
│ │ new_aan_pairs.test.tsv
│ │ new_aan_pairs.train.tsv
│ │
│ ├───IMDB
│ │ │ dev_contrast.tsv
│ │ │ dev_original.tsv
│ │ │ test_contrast.tsv
│ │ │ test_counterfactual.tsv
│ │ │
│ │ └───aclImdb
│ │ │ imdb.vocab
│ │ │ imdbEr.txt
│ │ │ README
│ │ │
│ │ ├───test
│ │ │ │ labeledBow.feat
│ │ │ │ urls_neg.txt
│ │ │ │ urls_pos.txt
│ │ │ │
│ │ │ ├───neg
│ │ │ └───pos
│ │ └───train
│ │ │ labeledBow.feat
│ │ │ unsupBow.feat
│ │ │ urls_neg.txt
│ │ │ urls_pos.txt
│ │ │ urls_unsup.txt
│ │ │
│ │ ├───neg
│ │ ├───pos
│ │ └───unsup
│ ├───listops
│ │ base.py
│ │ basic_test.tsv
│ │ dev_d7s.tsv
│ │ load_listops_data.py
│ │ make_depth_dev_data.py
│ │ make_depth_ndr_data.py
│ │ make_depth_test_data.py
│ │ make_depth_train_data.py
│ │ make_depth_train_data_extra.py
│ │ make_iid_data.py
│ │ make_odd_25depth_data.py
│ │ make_ood_10arg_data.py
│ │ make_ood_15arg_data.py
│ │ test_200_300.tsv
│ │ test_300_400.tsv
│ │ test_400_500.tsv
│ │ test_500_600.tsv
│ │ test_600_700.tsv
│ │ test_700_800.tsv
│ │ test_800_900.tsv
│ │ test_900_1000.tsv
│ │ test_d20s.tsv
│ │ test_dg8s.tsv
│ │ test_iid_arg.tsv
│ │ test_ood_10arg.tsv
│ │ test_ood_15arg.tsv
│ │ train_d20s.tsv
│ │ train_d6s.tsv
│ │ __init__.py
│ │
│ ├───listops_lra
│ │ basic_test.tsv
│ │ basic_train.tsv
│ │ basic_val.tsv
│ │
│ ├───MNLI
│ │ │ conj_dev.tsv
│ │ │ multinli_0.9_test_matched_unlabeled.jsonl
│ │ │ multinli_0.9_test_matched_unlabeled_hard.jsonl
│ │ │ multinli_0.9_test_mismatched_unlabeled.jsonl
│ │ │ multinli_0.9_test_mismatched_unlabeled.jsonl.zip
│ │ │ multinli_0.9_test_mismatched_unlabeled_hard.jsonl
│ │ │ multinli_1.0_dev_matched.jsonl
│ │ │ multinli_1.0_dev_matched.txt
│ │ │ multinli_1.0_dev_mismatched.jsonl
│ │ │ multinli_1.0_dev_mismatched.txt
│ │ │ multinli_1.0_train.jsonl
│ │ │ multinli_1.0_train.txt
│ │ │ paper.pdf
│ │ │ README.txt
│ │ │
│ │ ├───Antonym
│ │ │ multinli_0.9_antonym_matched.jsonl
│ │ │ multinli_0.9_antonym_matched.txt
│ │ │ multinli_0.9_antonym_mismatched.jsonl
│ │ │ multinli_0.9_antonym_mismatched.txt
│ │ │
│ │ ├───Length_Mismatch
│ │ │ multinli_0.9_length_mismatch_matched.jsonl
│ │ │ multinli_0.9_length_mismatch_matched.txt
│ │ │ multinli_0.9_length_mismatch_mismatched.jsonl
│ │ │ multinli_0.9_length_mismatch_mismatched.txt
│ │ │
│ │ ├───Negation
│ │ │ multinli_0.9_negation_matched.jsonl
│ │ │ multinli_0.9_negation_matched.txt
│ │ │ multinli_0.9_negation_mismatched.jsonl
│ │ │ multinli_0.9_negation_mismatched.txt
│ │ │
│ │ ├───Numerical_Reasoning
│ │ │ .DS_Store
│ │ │ multinli_0.9_quant_hard.jsonl
│ │ │ multinli_0.9_quant_hard.txt
│ │ │
│ │ ├───Spelling_Error
│ │ │ multinli_0.9_dev_gram_contentword_swap_perturbed_matched.jsonl
│ │ │ multinli_0.9_dev_gram_contentword_swap_perturbed_matched.txt
│ │ │ multinli_0.9_dev_gram_contentword_swap_perturbed_mismatched.jsonl
│ │ │ multinli_0.9_dev_gram_contentword_swap_perturbed_mismatched.txt
│ │ │ multinli_0.9_dev_gram_functionword_swap_perturbed_matched.jsonl
│ │ │ multinli_0.9_dev_gram_functionword_swap_perturbed_matched.txt
│ │ │ multinli_0.9_dev_gram_functionword_swap_perturbed_mismatched.jsonl
│ │ │ multinli_0.9_dev_gram_functionword_swap_perturbed_mismatched.txt
│ │ │ multinli_0.9_dev_gram_keyboard_matched.jsonl
│ │ │ multinli_0.9_dev_gram_keyboard_matched.txt
│ │ │ multinli_0.9_dev_gram_keyboard_mismatched.jsonl
│ │ │ multinli_0.9_dev_gram_keyboard_mismatched.txt
│ │ │ multinli_0.9_dev_gram_swap_matched.jsonl
│ │ │ multinli_0.9_dev_gram_swap_matched.txt
│ │ │ multinli_0.9_dev_gram_swap_mismatched.jsonl
│ │ │ multinli_0.9_dev_gram_swap_mismatched.txt
│ │ │
│ │ └───Word_Overlap
│ │ multinli_0.9_taut2_matched.jsonl
│ │ multinli_0.9_taut2_matched.txt
│ │ multinli_0.9_taut2_mismatched.jsonl
│ │ multinli_0.9_taut2_mismatched.txt
│ │
│ ├───proplogic
│ │ generate_neg_set_data.py
│ │ test0
│ │ test1
│ │ test10
│ │ test11
│ │ test12
│ │ test2
│ │ test3
│ │ test4
│ │ test5
│ │ test6
│ │ test7
│ │ test8
│ │ test9
│ │ train0
│ │ train1
│ │ train10
│ │ train11
│ │ train12
│ │ train2
│ │ train3
│ │ train4
│ │ train5
│ │ train6
│ │ train7
│ │ train8
│ │ train9
│ │ __init__.py
│ │
│ ├───QQP
│ │ │ quora_duplicate_questions.tsv
│ │ │
│ │ ├───PAWS_QQP
│ │ │ dev_and_test.tsv
│ │ │
│ │ └───PAWS_WIKI
│ │ test.tsv
│ │
│ └───SNLI
│ │ dataset.jsonl
│ │ README.txt
│ │ snli_1.0_dev.jsonl
│ │ snli_1.0_dev.txt
│ │ snli_1.0_test.jsonl
│ │ snli_1.0_test.txt
│ │ snli_1.0_test_hard.jsonl
│ │ snli_1.0_train.jsonl
│ │ snli_1.0_train.txt
│ │
│ └───revised_combined
│ test.tsv
├───embeddings
│ └───glove
│ glove.840B.300d.txt
├───processed_data
│ ├───AAN_lra
│ │ dev_normal.jsonl
│ │ metadata.pkl
│ │ test_normal.jsonl
│ │ train.jsonl
│ │
│ ├───IMDB
│ │ dev_normal.jsonl
│ │ metadata.pkl
│ │ test_contrast.jsonl
│ │ test_counterfactual.jsonl
│ │ test_normal.jsonl
│ │ test_original_of_contrast.jsonl
│ │ test_original_of_counterfactual.jsonl
│ │ train.jsonl
│ │
│ ├───IMDB_lra
│ │ dev_normal.jsonl
│ │ metadata.pkl
│ │ test_normal.jsonl
│ │ train.jsonl
│ │
│ ├───listops200speed
│ │ dev_normal.jsonl
│ │ metadata.pkl
│ │ test_normal.jsonl
│ │ train.jsonl
│ │
│ ├───listops500speed
│ │ dev_normal.jsonl
│ │ metadata.pkl
│ │ test_normal.jsonl
│ │ train.jsonl
│ │
│ ├───listops900speed
│ │ dev_normal.jsonl
│ │ metadata.pkl
│ │ test_normal.jsonl
│ │ train.jsonl
│ │
│ ├───listopsc
│ │ dev_normal.jsonl
│ │ metadata.pkl
│ │ test_200_300.jsonl
│ │ test_300_400.jsonl
│ │ test_400_500.jsonl
│ │ test_500_600.jsonl
│ │ test_600_700.jsonl
│ │ test_700_800.jsonl
│ │ test_800_900.jsonl
│ │ test_900_1000.jsonl
│ │ test_iid_arg.jsonl
│ │ test_LRA.jsonl
│ │ test_normal.jsonl
│ │ test_ood_10arg.jsonl
│ │ test_ood_15arg.jsonl
│ │ train.jsonl
│ │
│ ├───listopsmix
│ │ dev_normal.jsonl
│ │ metadata.pkl
│ │ test_200_300.jsonl
│ │ test_300_400.jsonl
│ │ test_400_500.jsonl
│ │ test_500_600.jsonl
│ │ test_600_700.jsonl
│ │ test_700_800.jsonl
│ │ test_800_900.jsonl
│ │ test_900_1000.jsonl
│ │ test_iid_arg.jsonl
│ │ test_LRA.jsonl
│ │ test_normal.jsonl
│ │ test_ood_10arg.jsonl
│ │ test_ood_15arg.jsonl
│ │ train.jsonl
│ │
│ ├───listops_lra
│ │ dev_normal.jsonl
│ │ metadata.pkl
│ │ test_normal.jsonl
│ │ train.jsonl
│ │
│ ├───listops_lra_speed
│ │ dev_normal.jsonl
│ │ metadata.pkl
│ │ test_normal.jsonl
│ │ train.jsonl
│ │
│ ├───MNLIdev
│ │ dev_normal.jsonl
│ │ metadata.pkl
│ │ test_antonym_matched.jsonl
│ │ test_antonym_mismatched.jsonl
│ │ test_conj_nli.jsonl
│ │ test_content_word_swap_matched.jsonl
│ │ test_content_word_swap_mismatched.jsonl
│ │ test_function_word_swap_matched.jsonl
│ │ test_function_word_swap_mismatched.jsonl
│ │ test_keyboard_swap_matched.jsonl
│ │ test_keyboard_swap_mismatched.jsonl
│ │ test_length_matched.jsonl
│ │ test_length_mismatched.jsonl
│ │ test_matched.jsonl
│ │ test_mismatched.jsonl
│ │ test_negation_matched.jsonl
│ │ test_negation_mismatched.jsonl
│ │ test_numerical.jsonl
│ │ test_swap_matched.jsonl
│ │ test_swap_mismatched.jsonl
│ │ test_word_overlap_matched.jsonl
│ │ test_word_overlap_mismatched.jsonl
│ │ train.jsonl
│ │
│ ├───proplogic
│ │ dev_normal.jsonl
│ │ metadata.pkl
│ │ test_0.jsonl
│ │ test_1.jsonl
│ │ test_10.jsonl
│ │ test_11.jsonl
│ │ test_12.jsonl
│ │ test_2.jsonl
│ │ test_3.jsonl
│ │ test_4.jsonl
│ │ test_5.jsonl
│ │ test_6.jsonl
│ │ test_7.jsonl
│ │ test_8.jsonl
│ │ test_9.jsonl
│ │ train.jsonl
│ │
│ ├───QQP
│ │ dev_normal.jsonl
│ │ metadata.pkl
│ │ test_normal.jsonl
│ │ test_PAWS_QQP.jsonl
│ │ test_PAWS_WIKI.jsonl
│ │ train.jsonl
│ │
│ ├───SNLI
│ │ dev_normal.jsonl
│ │ metadata.pkl
│ │ SNLI_dev_normal.jsonl
│ │ SNLI_metadata.pkl
│ │ SNLI_test_hard.jsonl
│ │ SNLI_test_normal.jsonl
│ │ SNLI_train.jsonl
│ │ test_break.jsonl
│ │ test_counterfactual.jsonl
│ │ test_hard.jsonl
│ │ test_normal.jsonl
│ │ train.jsonl
│ │
│ ├───SST2
│ │ dev_normal.jsonl
│ │ metadata.pkl
│ │ test_normal.jsonl
│ │ train.jsonl
│ │
│ └───SST5
│ dev_normal.jsonl
│ metadata.pkl
│ test_normal.jsonl
│ train.jsonl
Train:
python trian.py --model=[insert model name] -- dataset=[insert dataset name] --times=[insert total runs] --device=[insert device name] --model_type=[classifier/sentence_pair/sentence_pair2]
- Check argparser.py for exact options.
- sentence_pair2 is used for sequence interaction models for sequence matching tasks (NLI, paraphrase detection), otherwise sequence_pair is used for model_type (if nothing about sequence interaction is explicitly mentioned in the paper then we are talking about a different paper).
- Generally we use total times as 3. For LRA we use 2.
The nomenclature in the codebase and in the paper are a bit different. We provide a mapping here of the form ([codebase model name] == [paper model name])
- CRvNN == CRvNN
- CRvNN_nohalt == CRvNN (during stress test)
- OM == OM
- GT_GRC == GT-GRC
- EGT_GRC == EGT-GRC
- BT_GRC == BT-GRC
- BT_GRC_OS == BT-GRC OS (also BT-GRC OneSoft)
- EBT_GRC == EBT-GRC
- EBT_GRC_noslice == EBT-GRC (-slice)
- EBT_GRC512 == EBT-GRC (512)
- EBT_GRC512_noslice == EBT-GRC (-slice,512)
- GAU_IN == GAU
- EGT_GAU_IN == EBT-GAU
- EBT_GAU_IN == EBT-GAU
- S4DStack == S4D
- BalancedTreeGRC == BBT-GRC
- HGRC == RIR-GRC
- HCRvNN == RIR-CRvNN
- HOM == RIR-OM
- HEBT_GRC == RIR-EBT-GRC
- HEBT_GRC_noSSM == RIR-EBT-GRC ($-$S4D)
- HEBT_GRC_noRBA == RIR-EBT-GRC ($-$Beam Align)
- HEBT_GRC_random == RIR-EBT-GRC ($+$Random Align)
- HEBT_GRC_small == RIR-EBT-GRC (beam 5)
- HEBT_GRC_chunk20 == RIR-EBT-GRC (chunk 20)
- HEBT_GRC_chunk10 == RIR-EBT-GRC (chunk 10)
- MEGA == MEGA
@InProceedings{pmlr-v139-chowdhury21a,
title = {Modeling Hierarchical Structures with Continuous Recursive Neural Networks},
author = {Chowdhury, Jishnu Ray and Caragea, Cornelia},
booktitle = {Proceedings of the 38th International Conference on Machine Learning},
pages = {1975--1988},
year = {2021},
editor = {Meila, Marina and Zhang, Tong},
volume = {139},
series = {Proceedings of Machine Learning Research},
month = {18--24 Jul},
publisher = {PMLR},
pdf = {http://proceedings.mlr.press/v139/chowdhury21a/chowdhury21a.pdf},
url = {http://proceedings.mlr.press/v139/chowdhury21a.html},
abstract = {Recursive Neural Networks (RvNNs), which compose sequences according to their underlying hierarchical syntactic structure, have performed well in several natural language processing tasks compared to similar models without structural biases. However, traditional RvNNs are incapable of inducing the latent structure in a plain text sequence on their own. Several extensions have been proposed to overcome this limitation. Nevertheless, these extensions tend to rely on surrogate gradients or reinforcement learning at the cost of higher bias or variance. In this work, we propose Continuous Recursive Neural Network (CRvNN) as a backpropagation-friendly alternative to address the aforementioned limitations. This is done by incorporating a continuous relaxation to the induced structure. We demonstrate that CRvNN achieves strong performance in challenging synthetic tasks such as logical inference (Bowman et al., 2015b) and ListOps (Nangia & Bowman, 2018). We also show that CRvNN performs comparably or better than prior latent structure models on real-world tasks such as sentiment analysis and natural language inference.}
}
@InProceedings{Chowdhury2023beam,
title = {Beam Tree Recursive Cells},
author = {Ray Chowdhury, Jishnu and Caragea, Cornelia},
booktitle = {Proceedings of the 40th International Conference on Machine Learning},
year = {2023}
}
@inproceedings{
chowdhury2023efficient,
title={Efficient Beam Tree Recursion},
author={Jishnu Ray Chowdhury and Cornelia Caragea},
booktitle={Thirty-seventh Conference on Neural Information Processing Systems},
year={2023},
url={https://openreview.net/forum?id=PR5znB6BZ2}
}
@inproceedings{
chowdhury2023recursion,
title={Recursion in Recursion: Two-Level Nested Recursion for Length Generalization with Scalability},
author={Jishnu Ray Chowdhury and Cornelia Caragea},
booktitle={Thirty-seventh Conference on Neural Information Processing Systems},
year={2023},
url={https://openreview.net/forum?id=o6yTKfdnbA}
}
For any issues contact: jishnu.ray.c@gmail.com