Intro#

is problem is classification or regression?#

demo: rain prediction
  • classification: the rain mesurement is a continuous value, for example from 0 to 500;

  • regression: the rain mesurement is a by ranges for city management : from 0-25 do nothing; for 25-200 do something; for 200-500 close the city…

train#

When training a model, we optimize it vs. a loss function

  • Measures the “deviation” / “distance” of the model output to an input from the ground truth data or other measure.

  • Serves the optimization (Minimization / Optimization) step of the model’s parameters.

  • Usually is smooth and differentiable (Sub Gradient).

  • Examples: MSE, MAE, Cross Entropy, …

eval#

When evaluating a model, we measure its performance using metrics / scores.

  • Measures the fit of the model to real world measures of its performance.

  • Serves the evaluation step of a trained model or the progress of the training

  • Has no limitations but being computable.

  • Examples: Accuracy, Number of outliers, AUC, …