Model validation: Baseline model

[PiotMik]

Perform performance evaluation of the baseline model.

[PiotMik]

For the baseline model, I went for validation on test set only, without performing cross-validation. This choice was motivated by the following arguments:

• The model is just a linear-boundary fitting algorithm, so with low inherent risk of overfitting to start with
• I used L2 regularization to control weights and prevent overfitting
• I monitored the differences between train_loss -test_loss and train_accuracy-test_accuracy during training - they stayed in line with each other during training phase
• I verified multiple performance metrics across train and test sets, which turned out indistinguishable

Based on that I concluded that the performance on train set and unseen test sets does not differ, hence there's no overfit and the model is well fitted.