This shows you the differences between two versions of the page.
Both sides previous revision Previous revision Next revision | Previous revision | ||
cs401r_w2016:lab13 [2016/02/08 23:16] admin |
cs401r_w2016:lab13 [2021/06/30 23:42] (current) |
||
---|---|---|---|
Line 28: | Line 28: | ||
For this lab, you will use the Old Faithful dataset, which you can download here: | For this lab, you will use the Old Faithful dataset, which you can download here: | ||
- | [[http://hatch.cs.byu.edu/courses/stat_ml/old_faithful.mat|Old Faithful dataset]] | + | [[https://www.dropbox.com/s/h8b67cg8wff7bg0/old_faithful.mat?dl=0|Old Faithful dataset]] |
+ | |||
+ | **The first thing you should is mean-center your data (ie, compute the mean of the data, then subtract that off from each datapoint).** (If you don't do this, you'll get zero probabilities for all of your responsibilities, given the initial conditions discussed below.) | ||
The equations for implementing the EM algorithm are given in MLAPP 11.4.2.2 - 11.4.2.3. | The equations for implementing the EM algorithm are given in MLAPP 11.4.2.2 - 11.4.2.3. | ||
Line 36: | Line 38: | ||
- Compute the responsibilities $r_{ik}$ (Eq. 11.27) | - Compute the responsibilities $r_{ik}$ (Eq. 11.27) | ||
- Update the mixing weights $\pi_k$ (Eq. 11.28) | - Update the mixing weights $\pi_k$ (Eq. 11.28) | ||
- | - Update the means $\mu_k$ (Eq. 11.31) | ||
- Update the covariances $\Sigma_k$ (Eq. 11.32) | - Update the covariances $\Sigma_k$ (Eq. 11.32) | ||
+ | - Update the means $\mu_k$ (Eq. 11.31) | ||
- | Now, repeat until convergence. | + | Now, repeat until convergence. Note that if you change the order of operations, you may get slightly difference convergences than the reference image. |
Since the EM algorithm is deterministic, and since precise initial conditions for your algorithm are given below, the progress of your algorithm should closely match the reference image shown above. | Since the EM algorithm is deterministic, and since precise initial conditions for your algorithm are given below, the progress of your algorithm should closely match the reference image shown above. |