Keras/TF implementation of AdamW, SGDW, NadamW, Warm Restarts, and Learning Rate multipliers
FEATURES:
kernel_regularizer=l2(1e-4)
pip install keras-adamw
BREAKING:
utils_common.py
BUGFIXES:
fill_dict_in_order
: _list_of_vals
-> values_list
example.py
: from keras_adamw.optimizers
-> from keras_adamw
MISC:
New features:
from keras_adamw import
for both the optimizers and utils, only requiring import os; os.environ["TF_KERAS"]='1'
if using tensorflow.keras
imports.Breaking changes:
'1'
and '0'
, instead of 'True'
and 'False'
Bug fixes:
get_configs()
incorrectly cast eta_min
, eta_max
, and eta_t
via int
- changed to float
New features:
keras
+ tf.keras
)keras
+ tf.keras
)^total_iterations_wd
)utils
now contain all common optimizer ops, e.g. _apply_weight_decays()
tf.python.keras
, TF2 + Keras 2.3.0; see docstring^ - also compatible w/ TensorFlow 1.13.0 & 1.15.0, Keras 2.2.3-2.2.4
BUGFIXES:
lr_multipliers
were not being applied for AdamW
SGDW
was missing K.symbolic
wrapperand total_iterations != 0
was missing for NadamW
and SGDW
, yielding NaN optimizer weights for total_iterations = 0
Misc changes:
README.md
w/ info on total_iterations_wd
README.md
to keras_adamw
on module version selectiontest.sh
; each test has a dedicated folder to ease coverage isolation.travis.yml
:
TF_KERAS
, TF_EAGER
, TF_VERSION
, and KERAS_VERSION
flagsKERAS
flag; repo is based on Kerassetuptools
& conda update -all
commandsrequirements.txt
to requirements-test.txt
Initial release - includes: