labeled inverse cell frequency: IGM

idf_igm(expr, features = NULL, label, lambda = 7, thres = 0)

Arguments

expr: a matrix, features in row and cells in column
features: vector, feature names or indexes to compute
label: vector, group label of each cell
lambda: numeric, hyperparameter for IGM
thres: numeric, cell only counts when expr > threshold, default 0

Value

a vector of inverse gravity moment score for each feature

Details

$$\mathbf{IGM_i} = log(1+\lambda\frac{max(n_{i,j\in D})_{k}}{\sum_{k}^{K}((n_{i,j\in D})_{k}*r_{k})+e^{-8}})$$ where $\lambda$ is the hyper parameter, $n_{i,j\in D}$ is the number of cells containing feature $i$ in class $D$, $r_k$ is the rank of $n_{i,j\in D}$.

Examples

data <- matrix(rpois(100, 2), 10, dimnames = list(1:10))
smartid:::idf_igm(data, label = sample(c("A", "B"), 10, replace = TRUE))
#>        1        2        3        4        5        6        7        8 
#> 1.203973 1.203973 1.335001 1.203973 1.265666 1.203973 1.203973 1.335001 
#>        9       10 
#> 1.203973 1.203973