Generalization and Sampling

Learn how to

• Assess if your model is overfitting
• Gauge when to stop model training
• Create repeatable training, evaluation, and test datasets
• Establish performance benchmarks

Loss Metrics

• MSE = Mean Squared Error
• RMSE = Root Mean Squared Error

Summary

• Taking a derivative of our loss services as our guide towards a minima
• We could have more than one minima for complex services
• Loss functions
  • RMSE for regression problems
  • Cross entropy for classification

https://machinelearningmastery.com/cross-entropy-for-machine-learning

• Perfectly accurate model with an RMSE of zero, can perform badly against a set of new data that it had not seen before
• Generalization
• Data preparation
  • training
  • evaluation
  • testing
• Overfitting
• underfitting
•