05. Generalization

Generalization

---

Topic: Generalization

Course: GMLC

Date: 16 February 2019

Professor: Not specified

---

Resources

---

• https://developers.google.com/machine-learning/crash-course/generalization/video-lecture

• https://developers.google.com/machine-learning/crash-course/generalization/peril-of-overfitting

Key Points

---

• Generalization - model’s ability to adapt to new data (using the initial distribution)

• Overfitting  - what happens when a model is complex and tries to fit training data so closely, that it fails to adapt to new data

• Training set  - dataset used to train model

• Test set

  • a dataset used to test model after initial training or validation sets

  • Must be large enough

  • Must not be repeated (using the same test over and over)

• Prediction - model’s output when provided with an example

• Stationary - meaning that it doesn’t change within the data set

• Independently & Identically (i.i.d) - examples don’t influence each other, randomness of variables

Check your understanding

---

• Explain when does overfitting occur

• How do we train models to adapt properly to new unseen data? (Which 2 sets do we use)

• What are the qualities of a good test set?

• What do we mean by saying that examples are i.i.d?

Summary of Notes

---

• A good model is not too complex to cause overfitting (bad adaption to new data)

• Good model is created using 2 subsets - Training set & Test set

• Examples provided for test sets must be independent and identical