Given an input dataset, sample compositions that are consistent with the input. Specifically, this samples from a multinomial with mean \(\phi^{-1}(Bx)\). The default depth is 5e4. Modify the "depth" parameter to change this.
Usage
# S4 method for class 'lnm'
sample(x, size = 1, depth = 50000, newdata = NULL, ...)Arguments
- x
An object of class lnm with fitted parameters \(\hat{B}\) and which we want to use to form predictions on new samples.
- size
The number of samples to generate.
- depth
The depth to use when sampling the multinomial for each simulated element.
- newdata
New samples on which to form predictions. Defaults to NULL, in which case predictions are made at the same design points as those used during the original training.
- ...
Additional keyword arguments, for consistency with R's predict generic (never used).
Value
A matrix of dimension size x n_outcomes, where each row
represents one sample from the posterior predictive of the fitted
logistic-normal multinomial model. Each row sums up to the depth argument,
which defaults to 5e4.
Examples
example_data <- lnm_data(N = 50, K = 10)
xy <- dplyr::bind_cols(example_data[c("X", "y")])
fit <- lnm(
starts_with("y") ~ starts_with("x"), xy,
iter = 25, output_samples = 25
)
#> Chain 1: ------------------------------------------------------------
#> Chain 1: EXPERIMENTAL ALGORITHM:
#> Chain 1: This procedure has not been thoroughly tested and may be unstable
#> Chain 1: or buggy. The interface is subject to change.
#> Chain 1: ------------------------------------------------------------
#> Chain 1:
#> Chain 1:
#> Chain 1:
#> Chain 1: Gradient evaluation took 0.001756 seconds
#> Chain 1: 1000 transitions using 10 leapfrog steps per transition would take 17.56 seconds.
#> Chain 1: Adjust your expectations accordingly!
#> Chain 1:
#> Chain 1:
#> Chain 1: Begin eta adaptation.
#> Chain 1: Iteration: 1 / 250 [ 0%] (Adaptation)
#> Chain 1: Iteration: 50 / 250 [ 20%] (Adaptation)
#> Chain 1: Iteration: 100 / 250 [ 40%] (Adaptation)
#> Chain 1: Iteration: 150 / 250 [ 60%] (Adaptation)
#> Chain 1: Iteration: 200 / 250 [ 80%] (Adaptation)
#> Chain 1: Success! Found best value [eta = 1] earlier than expected.
#> Chain 1:
#> Chain 1: Begin stochastic gradient ascent.
#> Chain 1: iter ELBO delta_ELBO_mean delta_ELBO_med notes
#> Chain 1: Informational Message: The maximum number of iterations is reached! The algorithm may not have converged.
#> Chain 1: This variational approximation is not guaranteed to be meaningful.
#> Chain 1:
#> Chain 1: Drawing a sample of size 25 from the approximate posterior...
#> Chain 1: COMPLETED.
#> Warning: Pareto k diagnostic value is Inf. Resampling is disabled. Decreasing tol_rel_obj may help if variational algorithm has terminated prematurely. Otherwise consider using sampling instead.
head(sample(fit))
#> y1 y2 y3 y4 y5 y6 y7 y8 y9 y10
#> [1,] 104 451 2145 57 107 46157 182 152 375 270
#> [2,] 1 0 49378 0 5 609 0 2 4 1
#> [3,] 295 787 18146 799 9126 18705 1065 496 359 222
#> [4,] 7 50 44643 6 33 5224 15 0 10 12
#> [5,] 517 4354 13614 3405 422 25693 986 450 287 272
#> [6,] 1 20 42403 0 9 7557 9 0 0 1