Prediction

After training and validating a model, we often want to use models on unseen data. In this page, we describe the prediction functions available in ERM.

Using a trained model on new data

The most basic way to do prediction is to use predict.

predict(M, x)
````
The code above takes as input a model `M` and data point `x`, from which a prediction is
formed according to the model's parameters. You can also input a different choice of `theta`,
as in `predict(M, x, theta)`. By default, it is set to the optimal `theta` stored in `M`.

Of course, in many settings you may want to predict on a certain set of data.

For example the following two lines of code will allow you to compute predictions on the training
set. 

julia predict_y_from_train(M) predict_v_from_train(M)

These functions allow you to compute embedded and unembedded predictions (corresponding to `y` and `v`, respectively) on the
train set of `M`.

julia predict_y_from_test(M) predict_v_from_test(M)

These functions allow you to compute embedded and unembedded predictions (corresponding to `y` and `v`, respectively) on the
test set of `M`.

Additionally, if you would rather provide a single raw input `u`, we provide all the prediction functions
you could ever want. 

julia predict_y_from_u(M) predict_v_from_u(M)

## Recovering losses

You can compute the train and test losses using `trainloss(M)` and `testloss(M)`, respectively. 

You often also want to compute a confusion matrix when solving classification problems.

julia confusion_train(M) confusion_test(M) ``These two functions compute confusion matrices on the train and test sets (respectively) forM`.