fitDeconfounder: fit the deconfounder to estimate average treatment effect
In zhangly811/MvDeconfounder: Deconfounder

Description Usage Arguments

View source: R/main.R

fit the deconfounder to estimate average treatment effect

fitDeconfounder(
  data_dir,
  save_dir,
  factor_model,
  learning_rate = 1e-04,
  max_steps = 1e+05,
  latent_dim = 1,
  layer_dim = c(30, 10),
  batch_size = 1024,
  num_samples = 1,
  holdout_portion = 0.5,
  print_steps = 50,
  tolerance = 3,
  num_confounder_samples = 30,
  CV = 5,
  outcome_type = "linear"
)

`data_dir`	String: the directory where cohort data are stored
`save_dir`	String: the directory where results will be stored
`factor_model`	String: the type of probabilistic factor model to fit. Choices are: PMF or DEF.
`learning_rate`	Float: The learning rate for the probabilistic factor model.
`max_steps`	Integer: the maximum steps to run the probabilistic factor model.
`latent_dim`	Integer: the number of latent dimensions in PMF.
`layer_dim`	List: a list of length 2. The number of latent dimensions in each layer of the 2-layer DEF.
`batch_size`	Integer: the number of datapoints to use in each training step of the probabilistic model.
`num_samples`	Integer: number of samples from variational distribution used in updating variational parameters.
`holdout_portion`	Float: A value between 0 and 1. The proportion of data heldout for predictive model checking in checking the probabilistic model.
`print_steps`	Integer: Print the results during training.
`tolerance`	Integer: The termination criteria for training the probabilistic model.
`num_confounder_samples`	Integer: number of samples of substitute confounder from the posterior, input for the outcome model for estimating ATE.
`CV`	Integer: Fold of cross validation in the outcome model.
`outcome_type`	String: The type of outcome. Choices are: linear