------------------------------DATA on DRIVE--------------------------------------------------------------------------- contains 2 databases: data12000: 12 052 sequences: uniref database (UniProt Knowledgebase with 35% similarity) Max sequence length: 1500 Min sequence length: 20
data160000:
160 026 sequences: intersection of uniref database (UniProt Knowledgebase with 50% identity, 21 mil sequences) and
Swissprot (500 000 sequences) Max sequence length: 34348 Min sequence length: 6
both folders include 4 more txt files containing labeled 3gram sequences: perm1, perm2, perm3, perm4 perm1 -> original permutation perm2 -> permutation in all windows of size 5 perm3 -> permutation of random 3grams in windows of size 5 perm4 -> permutation of every third window of size 5
(for data160000 only perm4 for now, perm1,2,3 are on drive)
------------------------------NGRAM REPRESENTATIONS------------------------------------------------------------- contains models for 3gram representations: 3grams_vecdim100_ctx25.dict -> dictionary of 3grams 3grams_vecdim100_ctx25.word2vec -> word2vec model Glove/results -> Glove model
------------------------------SEQUENCE LEARNING------------------------------------------------------------------ sequence_representations.py -> script for generating learning ngram_representations.py -> script for getting ngram representations sequence_results: folder for log and model results (both log and model should containg folders Glove and word2vec)
------------------------------EXECUTION------------------------------------------------------------------ (in folder sequence_learning) python sequence_representations.py ../data/data160000/perm4 ../ngram_representations 3grams_vecdim100_ctx25 100 100 --nb_epoch 50 --Glove g --log_to_file l --activation tanh
(without --log_to_file argument all results are printed to the console) (other possible parameters are in the script)