uclnlp / jack

Jack the Reader
MIT License
258 stars 82 forks source link

[WIP] fix and refactor vocabulary-embedding foo #380

Closed dirkweissenborn closed 6 years ago

dirkweissenborn commented 6 years ago

Requires updating pretrained model links before merging

also refactored training and some tests.

Basically makes vocab and embeddings independent, which is also reflected in InputModules now. Embeddings are now also just added to SharedResources object, along with Vocab and Config.

codecov-io commented 6 years ago

Codecov Report

Merging #380 into master will increase coverage by 4.29%. The diff coverage is 64.8%.

Impacted file tree graph

@@            Coverage Diff            @@
##           master    #380      +/-   ##
=========================================
+ Coverage    48.1%   52.4%   +4.29%     
=========================================
  Files          91      89       -2     
  Lines        5216    5145      -71     
=========================================
+ Hits         2509    2696     +187     
+ Misses       2707    2449     -258
Impacted Files Coverage Δ
jack/util/tf/dropout.py 0% <ø> (ø)
jack/util/tf/simple.py 0% <ø> (ø)
jack/util/tf/masking.py 100% <ø> (ø)
jack/util/torch/segment.py 0% <ø> (ø)
jack/util/torch/xqa.py 0% <ø> (ø)
jack/util/tf/pairwise_losses.py 0% <ø> (ø)
jack/util/torch/rnn.py 0% <ø> (ø)
jack/util/tf/activations.py 50% <ø> (ø)
jack/util/torch/misc.py 0% <ø> (ø)
jack/util/tf/misc.py 100% <ø> (ø)
... and 44 more

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update 8438c57...156a1c3. Read the comment docs.