007-lec-002. training dataset, test dataset, validation dataset, online learing
@
How can you evaluate your predicting model?
We train predicting model with training dataset
And if we let model to predict with training dataset,
model must show 100% correct accuracy
This is not good way to evaluate model
@
So, generally, we devide entire dataset into training dataset and test dataset
More specifically, training dataset can be devided into trainging dataset
and validation dataset for tuning learning rate $$$\alpha$$$ and regularization strength $$$\lambda$$$
@
There is 'online learing' method
Suppose we have too big dataset
We devided it into small size dataset
And we pass each small size dataset into model
Point is result of training should remain in model