Multi-task learning - #7

Open

okyksl wants to merge 61 commits into

mainfrom

Commits on Dec 3, 2021

feat(mlflow): artifact logging
okyksl
committed
4e62012
feat(data): add maximum tokenizer length as an option
okyksl
committed
9bf8cd6
feat(data): add inference option
okyksl
committed
97730a6
feat(modeling): use a dynamically provided threshold for groups
okyksl
committed
1a54e3e
feat(mlflow): add mlflow prediction wrapper
okyksl
committed
b251fa8
feat(mlflow): add artifact for inference params
okyksl
committed
47adf42
fix(mlflow): use sigmoid on raw network scores
okyksl
committed
cceab6a
fix(mlflow): rename mlflow.py to infer.py
okyksl
committed
83f31ef
fix(data): prepare targets and groups
okyksl
committed
81c0c32
fix(data): provide targets in non-inference mode
okyksl
committed
4bf767c
docs(mlflow): add additional considerations to MLFlow deployment
okyksl
committed
29af38c
fix(data): add token types only if it is supported
okyksl
committed
9f339ac
feat(mlflow): add deployment option
okyksl
committed
dff1e5f
chore(): delete TODO file
okyksl
committed
fa86267
refactor(mlflow): use 'predictions' instead of 'logits'
okyksl
committed
44fe9d8
fix(mlflow): use artifact uri in logging model
okyksl
committed
71a2c63
fix(mlflow): use proper artifact and code paths
okyksl
committed
11d23ec
refactor(mlflow): log label artifacts with the model
okyksl
committed
287a9be
feat(data): preprocessing for offline testing format
okyksl
committed
8187edd
feat(sagemaker): add inference job using MLFlow 'log_model'
okyksl
committed
816d6dc
feat(runner): reload best model in training
okyksl
committed
85123b0
feat(mlflow): add tokenizer options as arguments to mlflow
okyksl
committed
3372819
fix(mlflow): access artifacts in the pyfunc context object
okyksl
committed
f4d6c1e
refactor(data): update filtering info message
okyksl
committed
bf39d2e
refactor(data): make 'tokenizer_max_len' optional
okyksl
committed
4f6a050
fix(data): use source length to return dataset length
okyksl
committed
7044ca9
fix(data): drop empty rows in preprocessing
okyksl
committed
fae90ac
fix(mlflow): cast predictions and probabilities to string
okyksl
committed
832492a
fix(sagemaker): use fixed 'source' field by renaming prior to job execution
okyksl
committed
e1e7d1b
chore(params): fix 'epoch' hyperparameter
okyksl
committed
c215e38
fix(data): apply literal_eval for filtering
okyksl
committed
7f58e71
feat(modeling): add focal loss-star
okyksl
committed
0490d45
fix(mlflow): upload full code path
okyksl
committed
fc03a59
fix(mlflow): add deployment dependencies
okyksl
committed
cf99e32
feat(data): compute target statistics
okyksl
committed
a29e13d
feat(data): support pickle, csv and excel dataframes
okyksl
committed
76f2ba2
feat(modeling): add inverse loss weighting
okyksl
committed
d5b35f2
feat(data): support `sectors` as a 1-head classification task
okyksl
committed
99a3b9b
docs(modeling): add documentation to multi task transformer
okyksl
committed
d7286f4
fix(modeling): return zero if coarse group is predicted as negative
okyksl
committed
44c9634
feat(modeling): add inverse loss weighting
okyksl
committed
eb36bdc
feat(mlflow): update model output format
okyksl
committed
51b0bb5
refactor(data): modularize dataset into text and target datasets
okyksl
committed
0cbfd3c
feat(data): support multi-task learning
okyksl
committed
23f0991
refactor(modeling): refactor and document modeling
okyksl
committed
dcbf81e
fix(eval): test on all samples
okyksl
committed
24f3cd5
fix(modeling): do not explicitly pass num_heads to models
okyksl
committed
3e95ce9
feat(sagemaker): use new data version
okyksl
committed
6a62c6d

Commits on Dec 5, 2021

feat(data): use new virtual analysis framework
okyksl
committed
a1e1426
feat(dataset): have an option to exclude unwanted target labels
okyksl
committed
37c7c67
fix(modeling): initialize module list correctly
okyksl
committed
46354e2
feat(runner): exclude "NOT_MAPPED" targets
okyksl
committed
ee3e89b
fix(runner): handle multi-task targets correctly
okyksl
committed
0d8c1b1
refactor(data): better error messages when groups and group_names are not in same length
okyksl
committed
8eb627b

Commits on Dec 15, 2021

fix(data): compute stats for multi-head dataset
okyksl
committed
0b6b1d7
refactor(modeling): rename MultiTargetHead to MultiTargetTransformer
okyksl
committed
90f4369
feat(data): add iterative option for labels
okyksl
committed
6e9c7f8
feat(eval): add multihead metrics
okyksl
committed
84a8fde
fix(modeling): work in iterative settings
okyksl
committed
3f50ca7
feat(trainer): multi-task trainer
okyksl
committed
32a5d7f

Commits on Dec 16, 2021

fix(runner): fix evaluation and training logic for multi-task learning
okyksl
committed
5c53dd7