=================================================================================

Validation Error is a measure of how well your model generalizes to new, unseen data (validation data). A large gap between training and validation errors often indicates overfitting, where the model is fitting the training data too closely and may not generalize well to new data.

As shown in Figure 3768a, in machine learning, more examples give your model a broader perspective and help it generalize better so that the variance is reduced. Figure 3806cb shows the effect of training data size on error. The training error and validation error can be used as proxies for bias and variance so that they can represent variance in machine learning.

Figure 3768a. Effect of training data size on error. (code)

============================================