wassname/Castor
1
0
Fork 0
mirror of https://github.com/wassname/Castor.git synced 2026-09-11 11:51:23 +08:00
Code Issues Packages Projects Releases Wiki Activity
105 Commits 6 Branches 0 Tags
lintool-patch-4
Commit Graph
105 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
Ralph Tang 7957dc7638 Tweak conv-rnn model (#75) 
* Tweak conv-rnn model

- Fix misplaced zero_grad()
- Tweak model hyperparams and optimization algorithm

* Fix typo

* Add new results

* Clean up extraneous code
 2017-10-29 15:55:36 -04:00
MeowFei 8fd1ddcb4c Added RetrieveSentences.py (#74) 
* added RetrieveSentences.py

* removed index
 2017-10-28 15:56:20 -04:00
MeowFei 00ef2d0fde added RetrieveSentences.py (#73) 2017-10-28 15:05:52 -04:00
rosequ 491f0b32d5 Updated sm (#72) 
+ removed redundant loss regularization
+ added script to create torch word embedding file from word2vec model
+ updated README
 2017-10-12 22:54:45 -04:00
Michael Tu 857cd43934 MP-CNN: Optionally Visualize Training for Debugging (#67) 
* MP-CNN: optionally support TensorBoard for learning curve visualization

* MP-CNN: bug with SummaryWriter comment

* MP-CNN: add instructions on how to setup tensorboard
 2017-10-05 19:15:51 -04:00
Michael Tu e61db8e7c0 Remove Kim CNN data (required for bfg repo cleaning) 2017-10-05 15:09:11 -04:00
Ralph Tang e7ca33de28 Add convolutional RNN for sentence classification (#57) 
* Add SST data preprocessing

* Add ConvRNN model

* Add LR scheduler

* Add grid search on hyperparameters

* Add random search

* Add CLI options

* Add usage to README.md

* Refactor code

* Fix randomized search parameters

* Update README.md with results

* Use Dataset and DataLoader
 2017-10-05 14:14:31 -04:00
Peng Shi 511f29a2a6 kim cnn with torchtext (#59) 
* kim cnn with torchtext

* model

* import style
 2017-10-05 09:38:07 -04:00
rosequ 36b1ddb858 SM model for WikiQA (#63) (#64) 
* support for WikiQA dataset

* parallel runs for both datasets

* minor fixes

* updated README

* removed data folder; added scripts to create dataset; updated README

* after CR

* after CR2
 2017-10-04 15:59:26 -04:00
rosequ a471cea2c5 reimplementation of SM model (#48) (#62) 
* reimplementation of SM model

* features without normalization; parallel running of different modes

* minor fix
 2017-10-03 15:02:34 -04:00
Michael Tu 6362239233 MP-CNN: early stopping (#60) 2017-09-27 19:26:20 -04:00
Michael Tu aba4dd7e89 MP-CNN: Add Additional Features (#51) 
* MP-CNN: support external features

* MP-CNN: calculate overlap in same way as SM-model

* MP-CNN: properly calculate idf overlap
 2017-09-21 22:43:31 -04:00
Matt Crane ed4dba2497 Typo fix (#54) 2017-09-22 11:30:26 +09:00
Matt Crane a601722768 Option to disable the CuDNN backend (#52) 
Some of the kernels in this backend are known to be non-deterministic. It's also not clear to me whether it's enabled in a default install or not, enabled is set to True by default, but I think ultimately it's a runtime choice.
 2017-09-22 11:19:30 +09:00
Michael Tu a2904efe5a MP-CNN Bugfixes and Improvements (#50) 
* MP-CNN: use consistent unknown vector

* MP-CNN: Make optimizer, patience, etc.. configurable

* MP-CNN: bug fixes

* MP-CNN: update README

* MP-CNN: remove unused import
 2017-09-21 15:53:54 -04:00
Michael Tu 245b032545 Set gpu seed for SM model (#46) 
Small fix to set GPU seed for SM model
 2017-09-12 14:28:18 -04:00
Matt Crane a3294339fb Add seed and thread arguments (#43) 
So that any experiments can be controlled from the command line, I've set the defaults to what they were hardcoded (for the seed) and num_threads in kim_cnn to be consistent with the default in sm_cnn.
 2017-09-01 09:30:01 -04:00
rosequ d2d958cc51 Kim CNN: Early Stop Training (#40) (#41) 
Stop training if there's no improvement in the accuracy over 5 epochs
 2017-08-30 14:45:01 -04:00
Jimmy Lin fc91a0b04a Nuking simple_qa_rnn since we have separate BuboQA repo. (#42) 
cf. https://github.com/castorini/BuboQA
 2017-08-29 14:21:09 -04:00
Matt Crane b4dc87d46f Explicitly open files with utf-8 encoding (#35) 
Same issue as castorini/data#19
 2017-08-29 14:20:18 -04:00
Michael Tu 449d715ab0 MP-CNN PyTorch Re-implementation (#37) 
Re-implementation of MP-CNN in PyTorch.
 2017-08-04 19:37:34 -04:00
Peng Shi 53e0de03d7 Kim's sentence classification model: Upgrade to Python 3 (#36) 
Kim's sentence classification model: Upgrade to Python 3, per issue #31
 2017-07-30 08:10:44 +08:00
Michael Tu a6fc10818f Fix SM Model Internal Reproducibility Bug (#34) 
* Fix SM Model reproducibility bug

vocab is in different order every time, causing unseen words to use
different random states

* Make requirements.txt usable from conda and pip

The existing torch requirement does not work with conda or pip.
Also upgrade pytorch version while we are at it.
 2017-07-20 06:52:26 +08:00
Salman Mohammed 061cc8dd09 Used torchtext to refactor relation prediction model - much cleaner! (#33) 
Refactored the code for Ferhan's relation prediction model using torchtext
 2017-07-15 07:06:10 +08:00
Michael Tu a0755e6aa3 SM Model Jupyter Notebook Tutorial (#32) 
Notebook tutorial for SM CNN.
 2017-07-04 18:28:17 -04:00
Salman Mohammed 4c081645a7 clean up the relation prediction model for Simple QA - Ferhan's paper (#28) 
+ cleaned up the model code for the simple qa directory
+ created vocab objects for pre-loading word embeddings easily
 2017-06-19 14:24:27 -04:00
Matt Crane 945b1fa6c0 Fix vocab caching issue (#29) 
With the line as-was the vocab cache was stored as b'the' rather than the, meaning that word2vec wasn't found for terms causing massive performance loss (AP 0.71 cf 0.77).
 2017-06-19 13:02:36 -04:00
Michael Tu 43dd6fdb1f GPU Support for SM Model (#26) 
Add code to using GPU for the SM Model (#25). To use the GPU if one is available, add the --cuda optional parameter when calling main.py.
 2017-06-02 08:47:13 -04:00
Gaurav Baruah 92789cb9f5 E2e sweep (#24) 
Now ensuring that the bridge process raw candidate sentences fetched from the index, exactly as was done for the best performing SM model.
 2017-05-29 19:47:42 -04:00
gauravbaruah 9bd5b7bb2a Corpus idf (#23) 
as part of sourcing-IDF-from-index and e2e experiments.
 2017-05-07 22:28:35 -04:00
rosequ 212aa6fb93 renaming sm_model; faster bridge (#18) (#22) 
+ renamed sm_model
+ faster bridge by obtaining the IDF scores of a term directly from the Java server
 2017-05-03 17:24:47 -04:00
Peng Shi 7d0a78d1a7 Kim cnn (#21) 
Reimplement Kim's sentence classification model #12
 2017-04-24 15:53:36 -04:00
Salman Mohammed d12a9cb475 Relation prediction model from Simple QA paper (#20) 
Initial implementation of RNNs for relation prediction described by Ture and Jojic:
https://arxiv.org/abs/1606.05029
 2017-04-20 13:29:32 -04:00
gauravbaruah a67e2d12c4 Ext feats bug fix (#19) 
+ sm model no external features baseline
+ sm model with IDF weights
+ sm model with IDF weights without removing punctuation --> barely better than df/idf (a la Pytorch).
+ sm model with stemming before computing IDF weights
^ all on the TrecQA dataset
 2017-04-18 12:32:43 -04:00
gauravbaruah 2758dee98f Idfbaselines (#17) 
IDF baselines:
+ using QA dataset only to compute IDF
+ using source corpus to compute IDF

Results are in baseline_results.tsv
 2017-04-16 20:29:13 -04:00
rosequ 3de103ac79 end2end qa pipeline (#16) 
+ Added a setup.py file
+ Reorganized the bridge file.
 2017-04-16 12:11:21 -04:00
rosequ b893f2a1b1 Idf baseline for QA datasets (#14) 
Added a functionality to generate idf scores for TrecQA and WikiQA datasets.
 2017-04-12 17:24:29 -04:00
gauravbaruah 906218b623 Improved README (#11) 
Issue #1 : Readme instructions updated for running the model
 2017-04-05 13:41:25 -04:00
gauravbaruah d4ac752cf2 e2e castorini/Castor castorini/data castorini/models (#7) 
Initial integration of Castor components for e2e QA demo.
 2017-04-02 15:24:33 -04:00
gauravbaruah 31928de93c Castorini smmodel now python 3 compatible. (#5) 
Model is now python 3.6 compatible.
 2017-03-31 19:36:11 -04:00
gauravbaruah 73577acc3f Castorini smmodel (#3) 
Code bridge between Castor/sm-model and Anserini
 2017-03-31 15:30:57 -04:00
gauravbaruah f91f105568 fixed variable perf bug and added debug arg for testing on each epoch (#2) 
Because of random word embedding for out of vocabulary words, the performance of the final saved model was variable. This is now fixed.
 2017-03-30 12:33:20 -04:00
Gaurav Baruah 67a93a736f important slight stopping criterion change 2017-03-28 15:03:22 -04:00
Gaurav Baruah b2c82f7e08 now preloading data 2017-03-28 14:30:48 -04:00
Gaurav Baruah e6aa9bfe07 added times to training 2017-03-28 13:37:41 -04:00
Gaurav Baruah ae77b6e57b fixed outputs, now stopping on MAP improvements 2017-03-28 13:03:20 -04:00
Gaurav Baruah 73ed76951c merged changes 2017-03-28 12:33:25 -04:00
Gaurav Baruah 7b20f50800 debugging for training loss 2017-03-28 12:01:48 -04:00
htaustin 46d913392b exp() preidction scores after logsoftmax 2017-03-27 22:48:35 -04:00
gauravbaruah c0f34a687e testing call fixed 2017-03-27 16:18:47 -04:00