Pro Machine Learning Algorithms by V Kishore Ayyadevara

Author:V Kishore Ayyadevara
Language: eng
Format: epub, pdf
Publisher: Apress, Berkeley, CA

Notice that, the above involves, invoking an additional hyper-parameter - “kernel_regularizer” and then specifying whether it is an L1 / L2 regularization. Further we also specify the value that gives the weightage to regularization.

We notice that, post regularization, training and test dataset accuracy are similar to each other, where training dataset accuracy is 97.6% while test dataset accuracy is 97.5%. The histogram of weights post L2 regularization is as follows:

We notice that a majority of weights are now much closer to 0 when compared to the previous two scenarios and thus avoiding the overfitting issue that was caused due to high weight values assigned for edge cases. We would see a similar trend in case of L1 regularization.

Thus, L1 and L2 regularizations help us in avoiding the issue of overfitting on top of training dataset but not generalizing on test dataset.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.