/my-model-trains

Silly library to regress your loss function and to feel confident about your training. We all know we cannot regress loss functions, right? But we do

Primary LanguagePythonGNU General Public License v3.0GPL-3.0

My model trains?

We know we can't but... we want. This package allows you to regress your loss function with 2 lines of code:

import my_model_trains as mmt

regression = mmt.plot_and_regress(np.arange(1, 1 + len(values)), values, 'SDR estimation 50 points', 'epochs', 'SDR',
                                  regressors=['natural_log', 'power_law'], verbose=True, n_values=1200)
plt.show()

Natural log: A = 1.8883062635915384, B = 1.7362624973685172, R^2 = 3.744796929596155
Power law: A = 3.169713596274487, B = 0.2575654072696419, R^2 = 7.63353926057481
Natural log: A = 1.191823982143853, B = 4.667963014944526, R^2 = 14.247145407884146
Power law: A = 5.316001879233005, B = 0.13215287499090037, R^2 = 19.166739574852727

imagen

Wooow! My metrics are gonna be soo good

Now we face the reality :D
imagen

In fact if we remove the first data points it's more accurate :)
imagen