estimate_timecourse_params_tf: Fit Timecourses

In calico/impulse: Fit Impulse and Sigmoidal Curves to Longitudinal Data Using TensorFlow

Fit Timecourses

Description

Fit Timecourses

Usage

estimate_timecourse_params_tf(
  measurements,
  model = "sigmoid",
  n_initializations = 100,
  max_n_initializations = 1000,
  use_prior = TRUE,
  prior_pars = c(v_sd = 1.2, rate_shape = 2, rate_scale = 0.25, time_shape = 2,
    time_scale = 15),
  fit_intercept = FALSE,
  learning_rate = 0.1,
  n_iterations = 100,
  verbose = FALSE
)

Arguments

measurements	a tibble containing: tc_id a unique indicator for each timecourse" time a numeric predictor variable abundance a numeric response variable noise (optional) if provided, a numeric variable specifying the noise estimate for each observation. This will be used to calculate the Gaussian likelihood if priors are used, or to calculate a weighted sum-of-squares estimate if priors are not used
model	model to fit: sigmoid one sigmoidal response impulse two sigmoidal responses
n_initializations	Number of initializations to use for each timecourse.
max_n_initializations	Maximum number of initializations that can be used
use_prior	If FALSE, fit least squares. If TRUE, add priors for a MAP estimate.
prior_pars	Named numeric vector of parameters to use for priors (if use_prior is TRUE) v_sd Gaussian SD of assymptotes (v_inter and v_final) rate_shape Shape of rate Gamma rate_scale Scale of rate Gamma time_shape Shape of t_rise and t_fall - t_rise Gamma time_scale Scale of t_rise and t_fall - t_rise Gamma
fit_intercept	If TRUE, the intercept will be fit, if FALSE, the intercept will be constrainted to zero
learning_rate	learning rate for the Adams optimizer
n_iterations	the number of iterations to run the optimizer
verbose	if TRUE then print additional information

Value

a timecourse list:

invalid_timecourse_fits

tibble of parameter initializations for initializations that went to NaN for debugging

loss

tibble of losses for each tc_id and inititalization (init_id)

parameters

tibble of final parameters for each tc_id and initialization (init_id)

Examples

library(dplyr)
library(impulse)
auto_config_tf()

set.seed(123)
timecourses <- simulate_timecourses(n = 2, observation_level_noise = 0.5)

timecourses %>%
  tidyr::unnest_legacy(measurements) %>%
  # separate by true model
  tidyr::nest_legacy(-true_model, .key = "measurements") %>%
  # fit all models to each timecourse
  tidyr::crossing(tibble::tibble(model = c("sigmoid", "impulse"))) %>%
  dplyr::mutate(timecourse_params = purrr::map2(
    measurements,
    model,
    estimate_timecourse_params_tf,
    n_initializations = 20,
    use_prior = TRUE,
    fit_intercept = TRUE
    ))

calico/impulse documentation built on June 4, 2024, 5:28 a.m.