Weightwatcher

Latest version: v0.7.5.2

Safety actively analyzes 629723 Python packages for vulnerabilities to keep your Python projects secure.

Scan your dependencies

Page 6 of 7

0.6.4.1

added plot esd5 to analyze() , plot of xmins vs alphas
added fit_entropy() metric to fit_powerlaw()

0.6.4

new fix_fingers='clip_xmax' implementation 160
num_fingers added to dataframe
added sigma to Log-log Plots 162
detX value is now on same scale as xmin
params no longer modified as side effect 156
more extensive unit tests
check longname now part of unit tests
distances NOT working yet in full
removed explicit tensorflow and torch dependences 154
Q in details data frame 147
new method signature: rescaled_evals, Wscale = RMT_Util.rescale_eigenvalues(evals)
removed telly

0.6.3.4

fixed bug in fix_fingers

0.6.3.3

Fixed unit tests, removed CKA and EUCLIDIAN tests for now
Note: min_evals default different for analyze() and describe()

0.6.3.1

issue 160: updated fix_fingers='clip_xmax', now only checks alpha decreases by 1 or more
issue 156: implemented everywhere: params = DEFAULT_PARAMS.copy()
added examples/WW-Albert.ipynb

0.6.3

removed explcit dependencies on tensorflow pytoch, and onnx
infers framework from model , then imports (tf | torch |onnx)
add ability to extract and read pytorch statedict files directly
removed combine_weights_and_bias as default, not implemented anywhere
fixed unit tests, moved model and watcher to object constructor

Page 6 of 7

© 2024 Safety CLI Cybersecurity Inc. All Rights Reserved.