Mainly for the sixth week Content machine learning application recommendations and system design.
What to do next
When training good one model, predicting unknown data discovery, how to improve it?
- Get more examples of training
- Try to reduce the number of features
- Try to get more features
- Try adding two-item features
- Try to reduce the degree of normalization λ
- Try to increase the degree of normalization λ
Do not rush to try these methods, but through a number of machine learning diagnostic methods to determine what the current algorithm is, which methods can improve the effectiveness of the algorithm, how to choose a more meaningful method.
?
How to evaluate a model
Diagnostic methods
A Deviations and variances
Two Normalization
Three Learning curve
Look back.
More data is good?
Error analysis
Precision & Recall
Coursera Machine Learning Notes (iv)