Commits

Commits on Aug 6, 2019

Removing print of "shuffle_batch/random_shuffle_queue_Size:0" , because this op does not exist in the new tf.data.Dataset reader
urialon
authored
8fb4c50
Copy full SHA for 8fb4c50

Commits on Jul 17, 2019

Merge pull request #27 from eladn/master
Show description for 01d1731
urialon
authored
01d1731
Copy full SHA for 01d1731
Merge remote-tracking branch 'upstream/master'
eladn
committed
64bebf7
Copy full SHA for 64bebf7
readme: remove paper-version note. We just added new functionalities. If new params not stated explicitly the default behavior haven't changed.
eladn
committed
1c92dab
Copy full SHA for 1c92dab
config: make logging go to stdout instead of stderr
eladn
committed
419c7eb
Copy full SHA for 419c7eb

Commits on Jul 16, 2019

keras & tf models: refactor - move aux classes below main model class
eladn
committed
86d5b43
Copy full SHA for 86d5b43
vocabs: fix save_to_file() to adapt to old format
eladn
committed
cbc6487
Copy full SHA for cbc6487
tf model: use logger instead of print()s
eladn
committed
b68f0e9
Copy full SHA for b68f0e9
vocabs: refactor error msg on failed load due to wrong min word idx
eladn
committed
cf716b5
Copy full SHA for cf716b5
config: rename param "NUM_BATCHES_TO_LOG_PROGRESS"
eladn
committed
dfba714
Copy full SHA for dfba714
vocabs: fix save&load to match old format (without special words); now old tf model file loads successfully
eladn
committed
41e13ec
Copy full SHA for 41e13ec

Commits on Jul 15, 2019

README: minor bold-updates order fix
eladn
authored
63a8019
Copy full SHA for 63a8019
add & support option `config.SEPARATE_OOV_AND_PAD` and `vocab.special_words`
eladn
committed
cb7df2b
Copy full SHA for cb7df2b
tensorflow model: make it work in TF2.0.0-beta1 (train+eval+predict+save+load) using tf.compat.v1 api
eladn
committed
1bc14d4
Copy full SHA for 1bc14d4
update version: TF2.0.0-alpha ==> TF2.0.0-beta1 (use tf.compat.v1.string_split())
eladn
committed
d43d0d0
Copy full SHA for d43d0d0
model base: create model save dir if not exists
eladn
committed
38ffab5
Copy full SHA for 38ffab5
README: add keras impl title, update requirements
eladn
committed
f710132
Copy full SHA for f710132
add requirements.txt
eladn
committed
0aa498e
Copy full SHA for 0aa498e

Commits on Jun 27, 2019

tensorflow model: print # trainable params
eladn
committed
49bbd64
Copy full SHA for 49bbd64

Commits on Jun 17, 2019

Update README with links to C and Python extractors
urialon
authored
dcdc3c3
Copy full SHA for dcdc3c3

Commits on Jun 3, 2019

keras model save fix (to also save vocabs)
eladn
committed
8242e6e
Copy full SHA for 8242e6e
model base tiny refactor
eladn
committed
6edde7a
Copy full SHA for 6edde7a
logger fix: don't print twice (turn off propagate)
eladn
committed
69b41b6
Copy full SHA for 69b41b6
keras model: refactor: add `_create_train_callbacks()`; add docs
eladn
committed
b79789e
Copy full SHA for b79789e
docs: add some docs to classes
eladn
committed
2547420
Copy full SHA for 2547420
keras modeL: save only when needed; adapt save+load+predict to TF2
eladn
committed
7da5dc4
Copy full SHA for 7da5dc4
vocabs: fix minor issue (field initialization position)
eladn
committed
cbe8fa6
Copy full SHA for cbe8fa6
reader: fix `process_input_row()` to adapt TF2
eladn
committed
df3b0ca
Copy full SHA for df3b0ca
config: refactor: add annotations + optional strings + `is_saving` + return None for non-relevant properties
eladn
committed
5891587
Copy full SHA for 5891587

Commits on Jun 2, 2019

migrate to TF-2.0.0-alpha; now keras model works with BS=1024 + TP=1200; use keras CB to perform logging and evaluate during training; move arg parsing to config; fix AttentionLayer so mask will be…
Show description for 46ae7ad
eladn
committed
46ae7ad
Copy full SHA for 46ae7ad
config: change default values
eladn
committed
365bc36
Copy full SHA for 365bc36
keras model: tiny fix
eladn
committed
27b7e22
Copy full SHA for 27b7e22
vocabs: add tiny doc line
eladn
committed
81d22bd
Copy full SHA for 81d22bd
main file minor refactor
eladn
committed
64f30d4
Copy full SHA for 64f30d4

Commits on May 27, 2019

keras model: refactor ckpt saver cb; refactor eval cb; cancel re-compile after ckpt restore
eladn
committed
97e752f
Copy full SHA for 97e752f