context("constraints")



test_constraint <- function(name) {
  constraint_fn <- eval(parse(text = name))
  test_call_succeeds(name, {
    keras_model_sequential() %>%
      layer_dense(32, input_shape = c(784),
                  kernel_constraint = constraint_fn(),
                  bias_constraint = constraint_fn())
  })
}


test_constraint("constraint_maxnorm")
test_constraint("constraint_minmaxnorm")
test_constraint("constraint_nonneg")
test_constraint("constraint_unitnorm")

test_succeeds("R custom constraints", {

  nonneg_constraint <- function(w) {
    # >= dispatches to op_greater_equal()
    w * op_cast(w >= 0, config_floatx())
  }

  constraint_custom_nonneg <- Constraint(
    "CustomNonNegConstraint",
    public = list(
      call = nonneg_constraint
    )
  )


  model <- keras_model_sequential(input_shape = c(784)) %>%
    layer_dense(32,
                kernel_constraint = constraint_custom_nonneg(),
                bias_constraint = nonneg_constraint) %>%
    layer_dense(10, activation = 'softmax') %>%
    compile(loss = 'binary_crossentropy',
            optimizer = optimizer_sgd(),
            metrics = 'accuracy')


  data <- matrix(rexp(1000 * 784), nrow = 1000, ncol = 784)
  labels <- matrix(round(runif(1000 * 10, min = 0, max = 9)), nrow = 1000, ncol = 10)

  model %>% fit(data, labels, verbose = 0, epochs = 2)
})