--- title: RegularizationCallback keywords: fastai sidebar: home_sidebar summary: "Perform Group Regularization in fastai Callback system" description: "Perform Group Regularization in fastai Callback system" nb_path: "nbs/07_regularizer.ipynb" ---
{% raw %}
{% endraw %} {% raw %}
 
{% endraw %} {% raw %}
{% endraw %} {% raw %}
learn = cnn_learner(dls, resnet18, metrics=accuracy)
learn.unfreeze()

learn.fit_one_cycle(3)
epoch train_loss valid_loss accuracy time
0 0.681536 0.466989 0.835589 00:11
1 0.358927 0.318825 0.865359 00:10
2 0.201207 0.220008 0.923545 00:10
{% endraw %} {% raw %}

class RegularizationCallback[source]

RegularizationCallback(granularity, wd=0.01) :: Callback

Basic class handling tweaks of the training loop by changing a Learner in various events

{% endraw %} {% raw %}
{% endraw %}

The RegularizationCallbackcan be used to perform $l_1$ regularization on any granularity available in the criteria class.

{% raw %}
reg_cb = RegularizationCallback('filter')
{% endraw %} {% raw %}
learn = cnn_learner(dls, resnet18, metrics=accuracy)
learn.unfreeze()

learn.fit_one_cycle(3, cbs=reg_cb)
epoch train_loss valid_loss accuracy time
0 1.633497 1.468702 0.812585 00:10
1 1.334702 1.173871 0.907307 00:10
2 1.152696 1.136654 0.933694 00:11
{% endraw %}