Learn R Programming

DeepLearningCausal (version 0.0.107)

deep_response_model: Response model from experimental data using deep neural learning through Tensorflow

Description

Train response model (response variable as outcome and covariates) from all compliers (actual + predicted) in experimental data using Tensorflow.

Usage

deep_response_model(
  response.formula,
  exp.data,
  exp.compliers,
  compl.var,
  algorithm = "adam",
  hidden.layer = c(2, 2),
  hidden_activation = "relu",
  epoch = 10,
  verbose = 1,
  batch_size = 32,
  output_units = 1,
  validation_split = NULL,
  patience = NULL,
  output_activation = "linear",
  loss = "mean_squared_error",
  metrics = "mean_squared_error",
  dropout_rate = NULL
)

Value

model object of trained response model.

Arguments

response.formula

formula specifying the response variable and covariates.

exp.data

experimental dataset.

exp.compliers

data.frame object of compliers from complier_predict.

compl.var

string specifying binary complier variable

algorithm

string for optimizer algorithm in response model.

hidden.layer

vector specifying hidden layers and the number of neurons in each hidden layer

hidden_activation

string or vector for activation functions in hidden layers.

epoch

integer for number of epochs

verbose

1 to display model training information and learning curve plot. 0 to suppress messages and plots.

batch_size

batch size to split training data.

output_units

integer for units in output layer. Defaults to 1 for continuous and binary outcome variables. In case of multinomial outcome variable, value should be set to the number of categories.

validation_split

double for the proportion of test data to be split as validation in response model.

patience

integer for number of epochs with no improvement after which training will be stopped.

output_activation

string for activation function in output layer. "linear" is recommended for continuous outcome variables, and "sigmoid" for binary outcome variables

loss

string for loss function. "mean_squared_error" recommended for linear models, "binary_crossentropy" for binary models.

metrics

string for metrics. "mean_squared_error" recommended for linear models, "binary_accuracy" for binary models.

dropout_rate

double or vector for proportion of hidden layer to drop out in response model.