Hands-On Machine Learning with TensorFlow.js

Fitting the curve

The final curve that's expressed by the final 2-degree polynomial model is as follows:

Unfortunately, it is difficult to say that our model fits the sine curve well. The training process seemed to work well because the loss function was constantly decreasing. There's nothing wrong in the training process itself. The problem is our choice of model. The sine curve is too complex to express with a 2-degree simple polynomial model. The model doesn't have sufficient expressiveness to represent the given sine curve. We call this problem underfitting with the high biased model. Underfitting is the situation when the algorithm oversimplifies ...