README.Rmd

---
output: github_document
---

<!-- README.md is generated from README.Rmd. Please edit that file -->

```{r, include = FALSE}
knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "man/figures/README-",
  out.width = "100%"
)
```

# SmoothHOOI

<!-- badges: start -->

<!-- badges: end -->

SmoothHOOI is an R package that implements methods described in "Smooth Tensor Decomposition with Application to Ambulatory Blood Pressure Monitoring Data". The application to real and synthetic ABPM data can be found [here](https://github.com/IrinaStatsLab/SmoothTensorDecompositionABPM).

## Installation

You can install the development version of SmoothHOOI from [GitHub](https://github.com/) with:

``` r
# install.packages("devtools")
devtools::install_github("IrinaStatsLab/SmoothHOOI")
```

## Example

This is a simple example which shows you how to use this package.

```{r}
library(SmoothHOOI)

set.seed(1234321)

# Generate a random tensor with missing data, in the array format
dims <- c(24, 3, 207)
tnsr <- array(rnorm(prod(dims), mean = 0, sd = 1), dim = dims)

missing_prob <- 0.2
missing_indices <- sample(length(tnsr), size = floor(length(tnsr) * missing_prob))
tnsr[missing_indices] <- NA
```

```{r}
# Make the second order difference matrix with circular nature
D2 <- SecDiffMat(24)
```

```{r}
# Find optimal hyperparameter with 5-fold cross-validation
kcv_res <- kcv(tnsr, rank_grid=as.matrix(expand.grid(r1<-seq(3,6,by=1), r2<-c(2,3))), lambda_seq=seq(1,10,by=1), k=5, L0=NULL, D=D2, tol=0.01, max_iter=500, init=0)

kcv_res$MSE_mat # matrix for CV error (rows representing ranks, cols representing lambda)
kcv_res$opt_para # optimal hyperparameters
```

```{r}
# Run SmoothHOOI algorithm with the optimal hyperparameters
res <- mglram(tnsr, ranks = c(3, 2), init=0, D = D2,
       lambda = 10, max_iter = 500, tol = 1e-5, L0 = NULL)

res$conv # check convergence
```

```{r}
# Rotation for Identifiability
tilde <- MakeIdent(L=res$L, G=res$G, R=res$R)

tilde$L_tilde
tilde$R_tilde
tilde$G_tilde[ , ,1]
```