A Metric
object encapsulates metric logic and state that can be used to
track model performance during training. It is what is returned by the family
of metric functions that start with prefix metric_*
.
A (subclassed) Metric
instance that can be passed directly to
compile(metrics = )
, or used as a standalone object. See ?Metric
for
example usage.
(Optional) string name of the metric instance.
(Optional) data type of the metric result.
model %>% compile(
optimizer = 'sgd',
loss = 'mse',
metrics = list(metric_SOME_METRIC(), metric_SOME_OTHER_METRIC())
)
m <- metric_SOME_METRIC()
for (e in seq(epochs)) {
for (i in seq(train_steps)) {
c(y_true, y_pred, sample_weight = NULL) %<-% ...
m$update_state(y_true, y_pred, sample_weight)
}
cat('Final epoch result: ', as.numeric(m$result()), "\n")
m$reset_state()
}
To be implemented by subclasses:
initialize()
: All state variables should be created in this method by calling self$add_weight()
like:
self$var <- self$add_weight(...)
update_state()
: Has all updates to the state variables like:
self$var$assign_add(...)
result()
: Computes and returns a value for the metric from the state variables.
Example custom metric subclass:
metric_binary_true_positives <- new_metric_class(
classname = "BinaryTruePositives",
initialize = function(name = 'binary_true_positives', ...) {
super$initialize(name = name, ...)
self$true_positives <-
self$add_weight(name = 'tp', initializer = 'zeros')
}, update_state = function(y_true, y_pred, sample_weight = NULL) {
y_true <- k_cast(y_true, "bool")
y_pred <- k_cast(y_pred, "bool")
values <- y_true & y_pred
values <- k_cast(values, self$dtype)
if (!is.null(sample_weight)) {
sample_weight <- k_cast(sample_weight, self$dtype)
sample_weight <- tf$broadcast_to(sample_weight, values$shape)
values <- values * sample_weight
}
self$true_positives$assign_add(tf$reduce_sum(values))
},
result = function()
self$true_positives
)
model %>% compile(..., metrics = list(metric_binary_true_positives()))
The same metric_binary_true_positives
could be built with %py_class%
like
this:
metric_binary_true_positives(keras$metrics$Metric) %py_class% {
initialize <- <same-as-above>,
update_state <- <same-as-above>,
result <- <same-as-above>
}