Skip to content

skallumadi/BachPropagate

BranchesTags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

25 Commits

data

data

 
 

notebooks

notebooks

 
 

.gitignore

.gitignore

 
 

LICENSE

LICENSE

 
 

README.md

README.md

 
 

build_bipartites_challenge.py

build_bipartites_challenge.py

 
 

build_bipartites_train.py

build_bipartites_train.py

 
 

build_indri_index_challenge.py

build_indri_index_challenge.py

 
 

build_indri_index_train.py

build_indri_index_train.py

 
 

emojiWords.txt

emojiWords.txt

 
 

requirements.txt

requirements.txt

 
 

stopwords.txt

stopwords.txt

 
 

utils.py

utils.py

 
 

utils_text.py

utils_text.py

 
 

Repository files navigation

Steps for training and testing and running on the challenge set

  1. Get track background data from spotify, create necessary pkls
    1. for the whole dataset for challenge run: python build_bipartites_challenge.py
  2. Create splits:
    1. BG750
    2. train
    3. test
    4. challenge(provided)
    5. telescoping
  3. Generate pkl where necessary-> pkl for trackfeatures and details; pid track mapping; splits pkls; most popular tracks
  4. clean the titles, *provide stop list *provide synonyms list, bigrams
  5. create background index documents, script to generate docs; config file to create indexes. 3 indexes -> Meta1, Meta2 and PRFQE
    1. run python build_indri_index_challenge.py
  6. create queries for these indexes for train, test and challenge
  7. script to generate results for the queries
  8. script to parse these results 2 different formats(BM25 vs QE)
  9. script to generate various metapaths and w2v models on BG 750 playlists;4 CBOW models ->a)just playlists, b)playlists and titles interspersed, c)AILA, d)ILI
  10. for each query playlist in splits, get 1000 items for the plalyist representation; batch generate and save
  11. main track: generate training data [PRFQE+BM25-1+BM25-2 -> 2000 items]+trackfeatures+playlist features+track-playlistfeatures+w2v features
  12. main track: generate test data(and challenge data)
  13. param sweep script to find right lambda mart model [train, test and evaluate]
  14. creative track: generate training data [PRFQE+BM25-1+BM25-2 -> 2000 items]+trackfeatures+playlist features+track-playlistfeatures+w2v features+track features from spotify api --> -missingzero
  15. run built model on challenge data
  16. prepare submission(popular items when playlist size <500)

About

Scripts/ code for Recsys Challenge 2018

Resources

Readme

License

Apache-2.0 license
Activity

Stars

7 stars

Watchers

3 watching

Forks

3 forks
Report repository

Releases

No releases published

Packages

No packages published

Languages