Introduction¶
Data Modeling¶
Data modeling generally broken down into
_____________
| 1.Data | ________________________ ______________________
‾‾‾‾‾‾‾‾‾‾‾‾‾ -> |3.Infer hidden variables| -> |4.Predict and explore |
_____________ ‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾ ‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾
|2.Build Model|
‾‾‾‾‾‾‾‾‾‾‾‾‾‾
Example - Guassian Multivariate¶
1.Data Data $\( x_{1}\ldots x_{n}|x_{i}\in \mathbb{R} ^{d} \)$
2. Build Model In this case we choose a probabilistic model Gaussian. We know the distribution family but don’t know the parameters for the model
We make an assumption about the data that it is i.i.d. (independant and identically distributed) $\( \chi _{i}n^{iid}p\left( x|\theta \right) ,i=1,\ldots n \)$
With the assumption that the data is iid when know that the joint probability distribution is equal to the product of the marginal distributions:x
3.Infer Hidden Variables Looking for the maxiumum likelihood we seek to find the value of \theta that maximizes the likelihood function
Logarithmic Trick Applying a log function does not change the location of the maximums of a function, just the value. If we apply the logarithm before deriving the gradient, it is easier to determine.