Andrew Landgraf

Add other exponential families

1

See table here: [http://en.wikipedia.org/wiki/Generalized_linear_model#Link_function](http://en.wikipedia.org/wiki/Generalized_linear_model#Link_function) - [ ] Exponential/Gamma distribution - [ ] Inverse Gaussian - [x] Categorical/Multinomial

Modify convergence criteria to percent change in loss. Need to determine what a good percent cutoff is.

Initialization does not take weights into account (unless there is normalization). Need to think of a general way to account for weights in the initialization. Maybe scale values by square...

Deviance doesn't monotonically decrease for gaussian with missing data

2

``` R rows = 100 cols = 10 set.seed(1) mat_np = outer(rnorm(rows), rnorm(cols)) mat = matrix(rpois(rows * cols, c(exp(mat_np))), rows, cols) missing_mat = matrix(runif(rows * cols)

Currently, the computational complexity is the same regardless of `majorizer`. The time per iteration can be sped up (to about `logisticPCA` speed) when `majorizer = "all"`.

- [ ] Missing data - [ ] Weights - [x] Gaussian

E.g. column 1 is continuous, column 2 is binary, column 3 is counts

Set it up to install libFM first, then do the tests on the package.

Add Tipping's formulation

1

Tipping, M. E. (1998). Probabilistic visualisation of high-dimensional binary data. NIPS 11, pp. 592-598.

I have written different method functions for lpca, lsvd, and clpca. I can probably combine many of them. The methods to combine are: - [ ] print - [ ]...