AAAI 2019 paper: "Document Informed Neural Autoregressive Topic Models with Distributional Prior" (a Contextualized Neural Topic Model with Word Embeddings)
This code consists of the implementations for the proposed models: DocNADE, iDocNADE, DocNADEe and iDocNADEe in the AAAI-19 paper.
Requires Python 3 (tested with 3.6.1). The remaining dependencies can then be installed via:
$ pip install -r requirements.txt
$ python -c "import nltk; nltk.download('all')"
Datasets: A directory containing CSV files. There is expected to be 1 CSV file per set or collection, with separate sets for training, validation and test. The CSV files in the directory must be named accordingly: training_docnade.csv, validation_docnade.csv, test_docnade.csv. For this task, each CSV file (prior to preprocessing) consists of 2 string fields with a comma delimiter - the first is the label and the second is the document body.
Vocabulary files: A plain text file, with 1 vocabulary token per line (note that this must be created in advance, we do not provide a script for creating vocabularies).
The script train_TMN_docnade_TASKTYPE.sh invokes train_model.py to train the four different model to compute PPL as well IR and save it in a repository. It will also log all the information with the PPL and IR in the same model folder. Here's how to use the script:
$ ./train_TMN_docnade_PPL.sh
$ ./train_TMN_docnade_IR.sh
- ``dataset`` is the path to the input dataset.
- ``docnadeVocab`` is the path to vocabulary file used by DocNADE.
- ``model`` is the path to model output directory.
- ``learning-rate`` is learning rate.
- ``batch-size`` is batch size for training data.
- ``num-steps`` is the number of steps to train for.
- `log-every`` is to print training loss after this many steps.
- ``validation-bs`` is the batch size for validation evaluation.
- ``test-bs`` is the batch size for test evaluation.
- ``validation-ppl-freq`` is to evaluate validation PPL and NLL after this many steps.
- ``validation-ir-freq`` is to evaluate validation IR after this many steps.
- ``test-ir-freq`` is to evaluate test IR after this many steps.
- ``test-ppl-freq`` is to evaluate test PPL and NLL after this many steps.
- ``num-classes`` is number of classes.
- ``patience`` is patience for early stopping criterion.
- ``hidden-size`` is size of the hidden layer.
- ``activation`` is which activation to use: sigmoid|tanh. Notice, use 'sigmoid' for **PPL** and 'tanh' for **IR** computations.
- ``bidirectional`` is whether to use iDocNADE model or not, i.e. True or False. If True, then model --> ``iDocNADE``
- ``initialize-docnade`` is whether to include glove embedding prior or not, i.e. True or False. If True, then model -->``DocNADEe``.
If ``bidirectional`` = True and ``initialize-docnade`` = True, then model --> ``iDocNADEe``.
- ``combination-type`` is combination type for iDocNADE forward and backward hidden document representation, for instance 'sum'
- ``vocab-size`` is the vocabulary size.
- ``lambda-embeddings`` is the mixture weight, i.e. [0.0-1.0] with word embeddings.
- ``projection`` is whether to project prior embeddings or not, i.e. True or False. Set to False.
- ``deep`` is whether to maked model deep (deepDocNADE) or not (docNADE/iDocNADE), i.e. True or False
- ``deep-hidden-sizes`` is sizes of the deep hidden layers for deepDocNADE, for instance, 200, 200.
- ``reload`` is whether to reload model or not, i.e. True or False
- ``reload-model-dir`` is path of directory for which model to be reloaded.
- ``trainfile`` is path to training text file (required in case of topic coherence), for instance, ./datasets/20NSshort/training.txt
- ``valfile`` is path to validation text file. (required in case of topic coherence), for instance, ./datasets/20NSshort/validation.txt
- ``testfile`` is path to test text file. (required in case of topic coherence), for instance, ./datasets/20NSshort/test.txt
- ``pretrained-embeddings-path`` is the path to pre-trained DocNADE model to initialize iDocNADE/DocNADEe/iDocNADEe model for **PPL** computation, for instance, './docnade_embeddings_ppl_reduced_vocab/TMNtitle'
Datasets directory: ./datasets/
Pre-trained embeddings dir: /home/usr/resources/pretrained_embeddings/
Results directory: ./model/MODELNAME/
Saved PPL model dir: ./model/MODELNAME/model_ppl/
Saved IR model dir: ./model/MODELNAME/model_ir/
Saved logs model dir: ./model/MODELNAME/logs/
Reload IR results: ./model/MODELNAME/logs/reload_info_ir.txt
Reload PPL results: ./model/MODELNAME/logs/reload_info_ppl.txt
--reload-model-dir
--trainfile
--valfile
--testfile
NOTE: In computing PPL or IR for larget text datasets, it is recommented to use the reload functionality for generating scores (PPL/IR) for the test set, and set test-ppl-freq or test-ir-freq to a very large number so as to avoid overhead during the training time.
@inproceedings{pankajgupta2018iDocNADEe, title={Document Informed Neural Autoregressive Topic Models with Distributional Prior}, author={Gupta, Pankaj and Chaudhary, Yatin and Buettner, Florian and Sch{"u}tze, Hinrich}, booktitle={AAAI}, year={2018} }
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent