Code for DeepXML: A Deep Extreme Multi-Label Learning Framework Applied to Short Text Documents

Last update: Nov 06, 2022

Overview

DeepXML

Code for DeepXML: A Deep Extreme Multi-Label Learning Framework Applied to Short Text Documents

Architectures and algorithms

DeepXML supports multiple feature architectures such as Bag-of-embedding/Astec, RNN, CNN etc. The code uses a json file to construct the feature architecture. Features could be computed using following encoders:

Bag-of-embedding/Astec: As used in the DeepXML paper [1].
RNN: RNN based sequential models. Support for RNN, GRU, and LSTM.
XML-CNN: CNN architecture as proposed in the XML-CNN paper [4].

Best Practices for features creation

Adding sub-words on top of unigrams to the vocabulary can help in training more accurate embeddings and classifiers.

Setting up

Expected directory structure

+-- 
   
    
|  +-- programs
|  |  +-- deepxml
|  |    +-- deepxml
|  +-- data
|    +-- 
    
     
|  +-- models
|  +-- results

Download data for Astec

* Download the (zipped file) BoW features from XML repository.  
* Extract the zipped file into data directory. 
* The following files should be available in 
   
    /data/
    
      for new datasets (ignore the next step)
    - trn_X_Xf.txt
    - trn_X_Y.txt
    - tst_X_Xf.txt
    - tst_X_Y.txt
    - fasttextB_embeddings_300d.npy or fasttextB_embeddings_512d.npy
* The following files should be available in 
     
      /data/
      
        if the dataset is in old format (please refer to next step to convert the data to new format)
    - train.txt
    - test.txt
    - fasttextB_embeddings_300d.npy or fasttextB_embeddings_512d.npy

Convert to new data format

# A perl script is provided (in deepxml/tools) to convert the data into new format as expected by Astec
# Either set the $data_dir variable to the data directory of a particular dataset or replace it with the path
perl convert_format.pl $data_dir/train.txt $data_dir/trn_X_Xf.txt $data_dir/trn_X_Y.txt
perl convert_format.pl $data_dir/test.txt $data_dir/tst_X_Xf.txt $data_dir/tst_X_Y.txt

Example use cases

A single learner with DeepXML framework

The DeepXML framework can be utilized as follows. A json file is used to specify architecture and other arguments. Please refer to the full documentation below for more details.

./run_main.sh 0 DeepXML EURLex-4K 0 108

An ensemble of multiple learners with DeepXML framework

An ensemble can be trained as follows. A json file is used to specify architecture and other arguments.

./run_main.sh 0 DeepXML EURLex-4K 0 108,666,786

Full Documentation

./run_main.sh 
    
     
      
       
       
         * gpu_id: Run the program on this GPU. * framework - DeepXML: Divides the XML problems in 4 modules as proposed in the paper. - DeepXML-OVA: Train the architecture in 1-vs-all fashion [4][5], i.e., loss is computed for each label in each iteration. - DeepXML-ANNS: Train the architecture using a label shortlist. Support is available for a fixed graph or periodic training of the ANNS graph. * dataset - Name of the dataset. - Astec expects the following files in 
        
         /data/
         
           - trn_X_Xf.txt - trn_X_Y.txt - tst_X_Xf.txt - tst_X_Y.txt - fasttextB_embeddings_300d.npy or fasttextB_embeddings_512d.npy - You can set the 'embedding_dims' in config file to switch between 300d and 512d embeddings. * version - different runs could be managed by version and seed. - models and results are stored with this argument. * seed - seed value as used by numpy and PyTorch. - an ensemble is learned if multiple comma separated values are passed.

Notes

* Other file formats such as npy, npz, pickle are also supported.
* Initializing with token embeddings (computed from FastText) leads to noticible accuracy gain in Astec. Please ensure that the token embedding file is available in data directory, if 'init=token_embeddings', otherwise it'll throw an error.
* Config files are made available in deepxml/configs/
   
    /
    
      for datasets in XC repository. You can use them when trying out Astec/DeepXML on new datasets.
* We conducted our experiments on a 24-core Intel Xeon 2.6 GHz machine with 440GB RAM with a single Nvidia P40 GPU. 128GB memory should suffice for most datasets.
* Astec make use of CPU (mainly for nmslib) as well as GPU.

Cite as

@InProceedings{Dahiya21,
    author = "Dahiya, K. and Saini, D. and Mittal, A. and Shaw, A. and Dave, K. and Soni, A. and Jain, H. and Agarwal, S. and Varma, M.",
    title = "DeepXML: A Deep Extreme Multi-Label Learning Framework Applied to Short Text Documents",
    booktitle = "Proceedings of the ACM International Conference on Web Search and Data Mining",
    month = "March",
    year = "2021"
}

References

[1] K. Dahiya, D. Saini, A. Mittal, A. Shaw, K. Dave, A. Soni, H. Jain, S. Agarwal, and M. Varma. Deepxml: A deep extreme multi-label learning framework applied to short text documents. In WSDM, 2021.

[2] pyxclib: https://github.com/kunaldahiya/pyxclib

[3] H. Jain, V. Balasubramanian, B. Chunduri and M. Varma, Slice: Scalable linear extreme classifiers trained on 100 million labels for related searches, In WSDM 2019.

[4] J. Liu, W.-C. Chang, Y. Wu and Y. Yang, XML-CNN: Deep Learning for Extreme Multi-label Text Classification, In SIGIR 2017.

[5] R. Babbar, and B. Schölkopf, DiSMEC - Distributed Sparse Machines for Extreme Multi-label Classification In WSDM, 2017.

[6] P., Bojanowski, E. Grave, A. Joulin, and T. Mikolov. Enriching word vectors with subword information. In TACL, 2017.

Code for DeepXML: A Deep Extreme Multi-Label Learning Framework Applied to Short Text Documents

Related tags

Overview

DeepXML

Architectures and algorithms

Best Practices for features creation

Setting up

Expected directory structure

Download data for Astec

Convert to new data format

Example use cases

A single learner with DeepXML framework

An ensemble of multiple learners with DeepXML framework

Full Documentation

Notes

Cite as

YOU MAY ALSO LIKE

References

Owner

Extreme Classification

Pytorch implementation of paper "Efficient Nearest Neighbor Language Models" (EMNLP 2021)

Diverse graph algorithms implemented using JGraphT library.

TensorFlow implementation of ENet

Hyperbolic Hierarchical Clustering.

Official code of CVPR 2021's PLOP: Learning without Forgetting for Continual Semantic Segmentation

[ICML 2020] "When Does Self-Supervision Help Graph Convolutional Networks?" by Yuning You, Tianlong Chen, Zhangyang Wang, Yang Shen

SuperSDR: multiplatform KiwiSDR + CAT transceiver integrator

Neural Reprojection Error: Merging Feature Learning and Camera Pose Estimation

GeneralOCR is open source Optical Character Recognition based on PyTorch.

WRENCH: Weak supeRvision bENCHmark

Official repository for "Restormer: Efficient Transformer for High-Resolution Image Restoration". SOTA for motion deblurring, image deraining, denoising (Gaussian/real data), and defocus deblurring.

Collections for the lasted paper about multi-view clustering methods (papers, codes)

Six - a Python 2 and 3 compatibility library

Data and analysis code for an MS on SK VOC genomes phenotyping/neutralisation assays

On Evaluation Metrics for Graph Generative Models

This repository compare a selfie with images from identity documents and response if the selfie match.

Metrics to evaluate quality and efficacy of synthetic datasets.

A rule learning algorithm for the deduction of syndrome definitions from time series data.

This repository contains a CBIR system that uses swin transformer to extract image's feature.

Complementary Patch for Weakly Supervised Semantic Segmentation, ICCV21 (poster)