Learning curves

A learning curve is a useful tool that displays how the validation and training score evolve as the number of training samples evolves.

The purpose of the learning curve is to find out whether and how much the model would benefit from using more data during training. It is also useful to diagnose whether the model's generalization error is more likely driven by bias or variance.

If, for example, both the validation score and the training score converge to a similarly low value despite an increasing training set size, the error is more likely due to bias, and additional training data is unlikely to help.

Take a look at the following visualization:

    ..................Content has been hidden....................

    You can't read the all page of ebook, please click here login for view all page.
    Reset
    52.91.84.219