TODO: Understand the expected value of the mean and variance
I am starting to see where the N - 1 comes from in the standard deviation calculation we use
TODO: also understand the cov calculation
What to do when uniform distribution is found in the data
Does polynomial catch the nn relations ?
Why did sqrt work better than log scale ?
- The more space the data occupies the easier the model will learn, because a simple differential change wouldnt change the target much [Elaborate More]
-
Variance in models: High variance: changing a small piece of the training data, would change the model structure or its parameters alot
A low variance model linear regression and logistic regression models for examle, they are robust for outliers. having an outlier wont change the slope of the line much, unlike tree models which are sensitive to outliers
Bias in models: Low bias: A model that is flexible and can correctly fit the data
High bias: models that are usually simple and cant fit the data, they have a constant sway from the data points (bias).
Stock price example p: 29 pdf
- Linear regression is linear in model parameters, not in the features