GithubHelp home page GithubHelp logo

secretpolice1995 / mspars_v2.0 Goto Github PK

View Code? Open in Web Editor NEW

This project forked from msra-nlc/mspars_v2.0

0.0 0.0 0.0 75.95 MB

License: Apache License 2.0

Python 85.84% Shell 3.63% Perl 4.93% Smalltalk 0.26% Emacs Lisp 2.30% JavaScript 0.11% NewLisp 0.21% Ruby 0.22% Slash 0.04% SystemVerilog 0.02% Makefile 0.04% Jupyter Notebook 0.89% TeX 1.51%

mspars_v2.0's Introduction

MSParS-V2.0

Code and dataset for paper "Asking Clarification Questions in Knowledge-Based Question Answering".

Folder Description

  1. data. It lists the constructed dataset, which is split into two parts: single-turn data and multi-turn data. Each part contains the data of three tasks: task1--clarification identification, task2--clarification generation, task3--question answering.
  2. code. The code folder includes two kinds of models: classification models and genertive models. Task1 and task3 share the same classification models while task2 uses generative models.

Task 1: Clarification Identification

Requirements

  • python 3.5.6
  • torchtext 0.4.0
  • torch 1.2.0
  • h5py 2.8.0
  • tflearn 0.3.2
  • pytorch 1.0.1

Baseline 1: CNN

Preprocessing

cd code/classification/preprocess_file
python3 pre_process.py --data_path ../../../data/multi-turn/task1/ --task task1 #Generate data for baseline CNN

Train & Test

cd ../a02_TextCNN/
python3 p7_TextCNN_train.py --cache_file_h5py ../../../data/multi-turn/task1/data.h51 --cache_file_pickle ../../../data/multi-turn/task1/vocab_label.pik #CNN

Baseline 2: HAN

Preprocessing

cd code/classification/preprocess_file
python3 pre_process_hir.py --data_path ../../../data/multi-turn/task1/ --task task1  #Generate data for HAN and DMN

Train & Test

cd ../a05_HierarchicalAttentionNetwork/
python3 p1_HierarchicalAttention_train.py --cache_file_h5py ../../../data/multi-turn/task1/hir_data.h5 --cache_file_pickle ../../../data/multi-turn/task1/hir_vocab_label.pik --learning_rate 0.005#HAN

Baseline 3: DMN

Preprocessing

cd code/classification/preprocess_file
python3 pre_process_predicate_hir.py --data_path ../../../data/multi-turn/task1/ --task task1  #Generate data for HAN and DMN

Train & Test

cd ../a09_DynamicMemoryNet/
python3 a8_train.py --cache_file_h5py ../../../data/multi-turn/task1/hir_data.h5 --cache_file_pickle ../../../data/multi-turn/task1/hir_vocab_label.pik --learning_rate 0.005  #DMN

Baseline 4: Transformer

Preprocessing

cd code/classification/preprocess_file
python3 generate_input_file.py --data_path ../../../data/multi-turn/task1/ --task task1  #Generate data for transformer and rnn
cd code/classification/sequence_model
python3 preprocess.py -train_src ../../../data/multi-turn/task1/src-train.txt -train_tgt ../../../data/multi-turn/task1/tgt-train.txt -valid_src ../../../data/multi-turn/task1/src-test.txt -valid_tgt ../../../data/multi-turn/task1/tgt-test.txt -save_data ../../../data/multi-turn/task1/demo -dynamic_dict

Train & Test

cd code/classification/sequence_model
python3 train.py -data ../../../data/multi-turn/task1/demo -save_model available_models/demo-model-transformer -gpu_ranks 0 -layers 4 -rnn_size 128 -word_vec_size 128 -transformer_ff 128 -heads 8  -encoder_type transformer -decoder_type transformer -position_encoding -dropout 0.1 -batch_size 64 -accum_count 2 -optim adam -adam_beta2 0.998 -decay_method noam -learning_rate 1 -max_grad_norm 0 -param_init 0  -param_init_glorot -label_smoothing 0.1 -valid_step 200

Task 2: Clarification Generation

Requirements

  • python 3.5.6
  • torchtext 0.4.0
  • torch 1.0.1

Preprocessing

cd code/generation
python3 preprocess.py -train_src ../../../data/multi-turn/task2/src-train.txt -train_tgt ../../../data/multi-turn/task2/tgt-train.txt -valid_src ../../../data/multi-turn/task2/src-test.txt -valid_tgt ../../../data/multi-turn/task2/tgt-test.txt -save_data ../../../data/multi-turn/task2/demo -dynamic_dict -share_vocab

Train

 python3 train.py -data ../../../data/multi-turn/task2/demo -save_model available_models/demo-model-transformer -gpu_ranks 0 -layers 1 -rnn_size 128 -word_vec_size 128 -transformer_ff 128 -heads 8  -encoder_type transformer -decoder_type transformer -position_encoding -dropout 0.1 -batch_size 16 -accum_count 2 -optim adam -adam_beta2 0.998 -decay_method noam -learning_rate 2 -max_grad_norm 0 -param_init 0  -param_init_glorot -label_smoothing 0.1 -valid_step 1000 -train_steps 50000 -save_checkpoint_steps 5000

Test

python3 translate.py -model available_models/demo-model-transformer_step_10000.pt -src ../../../data/multi-turn/task2/src-test.txt -output output.txt -replace_unk -verbose -gpu 0 -beam_size 1
python merge.py
perl tools/multi-bleu.perl ../../../data/multi-turn/task2/final_tgt_test.txt < final_output.txt

Task 3.1: Clarification-based Question Answering -- Predicate Prediction

Requirements

  • python 3.5.6
  • torchtext 0.4.0
  • torch 1.2.0
  • h5py 2.8.0
  • tflearn 0.3.2
  • pytorch 1.0.1

Baseline 1: CNN

Preprocessing

cd code/classification/preprocess_file
python3 pre_process_predicate_classifier.py --data_path ../../../data/multi-turn/task3/ --task task3.1 #Generate data for baseline CNN

Train & Test

cd ../a02_TextCNN/
python3 p7_TextCNN_train.py --cache_file_h5py ../../../data/multi-turn/task3/data.h51 --cache_file_pickle ../../../data/multi-turn/task3/vocab_label.pik #CNN

Baseline 2: HAN

Preprocessing

cd code/classification/preprocess_file
python3 pre_process_predicate_hir.py --data_path ../../../data/multi-turn/task3/ --task task3.1  #Generate data for HAN and DMN

Train & Test

cd ../a05_HierarchicalAttentionNetwork/
python3 p1_HierarchicalAttention_train.py --cache_file_h5py ../../../data/multi-turn/task3/hir_data.h5 --cache_file_pickle ../../../data/multi-turn/task3/hir_vocab_label.pik --num_classes 2 #HA

Baseline 3: DMN

Preprocessing

cd code/classification/preprocess_file
python3 pre_process_predicate_hir.py --data_path ../../../data/multi-turn/task3/ --task task3.1  #Generate data for HAN and DMN

Train & Test

cd ../a09_DynamicMemoryNet/
python3 a8_train.py --cache_file_h5py ../../../data/multi-turn/task3/hir_data.h5 --cache_file_pickle ../../../data/multi-turn/task3/hir_vocab_label.pik --num_classes 2 #DMN

Baseline 4: Transformer

Preprocessing

cd code/classification/preprocess_file
python3 generate_input_file_predicate.py --data_path ../../../data/multi-turn/task3/ --task task3.1  #Generate data for transformer and rnn
cd code/classification/sequence_model
python3 preprocess.py -train_src ../../../data/multi-turn/task3/src-train.txt -train_tgt ../../../data/multi-turn/task3/tgt-train.txt -valid_src ../../../data/multi-turn/task3/src-test.txt -valid_tgt ../../../data/multi-turn/task3/tgt-test.txt -save_data ../../../data/multi-turn/task3/demo -dynamic_dict

Train & Test

cd code/classification/sequence_model
python3 train.py -data ../../../data/multi-turn/task3/demo -save_model available_models/demo-model-transformer -gpu_ranks 0 -layers 1 -rnn_size 128 -word_vec_size 128 -transformer_ff 128 -heads 8  -encoder_type transformer -decoder_type transformer -position_encoding -dropout 0.1 -batch_size 8 -accum_count 2 -optim adam -adam_beta2 0.998 -decay_method noam -learning_rate 2 -max_grad_norm 0 -param_init 0  -param_init_glorot -label_smoothing 0.1 -valid_step 200

Task 3.2: Clarification-based Question Answering -- Entity Prediction

Requirements

  • python 3.5.6
  • torchtext 0.4.0
  • torch 1.2.0
  • h5py 2.8.0

Baseline 1: CNN

Preprocessing

cd code/classification/preprocess_file
python3 pre_process_entity_classifier.py --data_path "../../../data/multi-turn/task3/" --task "task3.2" #Generate data for baseline CNN

Train & Test

cd ../a02_TextCNN/
python3 p7_TextCNN_train.py --cache_file_h5py ../../../data/multi-turn/task3/data.h51 --cache_file_pickle ../../../data/multi-turn/task3/vocab_label.pik

Baseline 2: HAN

Preprocessing

cd code/classification/preprocess_file
python3 pre_process_hir_entity_classifier.py --data_pat "../../../data/multi-turn/task3/" --task "task3.2"  #Generate data for HAN and DMN

Train & Test

cd ../a05_HierarchicalAttentionNetwork/
python3 p1_HierarchicalAttention_train.py --cache_file_h5py ../../../data/multi-turn/task3/hir_data.h5 --cache_file_pickle ../../../data/multi-turn/task3/hir_vocab_label.pik #HAN

Baseline 3: DMN

Preprocessing

cd code/classification/preprocess_file
python3 pre_process_hir_entity_classifier.py "../../../data/multi-turn/task3/" --task "task3.2"  #Generate data for HAN and DMN

Train & Test

cd ../a09_DynamicMemoryNet/
python3 a8_train.py --cache_file_h5py ../../../data/multi-turn/task3/hir_data.h5 --cache_file_pickle ../../../data/multi-turn/task3/hir_vocab_label.pik #HAN

Baseline 4: Transformer

Preprocessing

cd code/classification/preprocess_file
python generate_input_file_entity.py --data_path "../../../data/multi-turn/task3/" --task "task3.2"
cd code/classification/sequence_model
python3 preprocess.py -train_src ../../../data/multi-turn/task3/src-train.txt -train_tgt ../../../data/multi-turn/task3/tgt-train.txt -valid_src ../../../data/multi-turn/task3/src-test.txt -valid_tgt ../../../data/multi-turn/task3/tgt-test.txt -save_data ../../../data/multi-turn/task3/demo-entity -dynamic_dict

Train & Test

cd code/classification/sequence_model
python3 train.py -data ../../../data/multi-turn/task3/demo -save_model available_models/demo-model-transformer -gpu_ranks 0 -layers 1 -rnn_size 128 -word_vec_size 128 -transformer_ff 128 -heads 8  -encoder_type transformer -decoder_type transformer -position_encoding -dropout 0.1 -batch_size 8 -accum_count 2 -optim adam -adam_beta2 0.998 -decay_method noam -learning_rate 2 -max_grad_norm 0 -param_init 0  -param_init_glorot -label_smoothing 0.1 -valid_step 200

News

  • Update and review single-turn data
  • Update and review multi-turn data
  • Update dataset split

mspars_v2.0's People

Contributors

jingjingxupku avatar

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.