% Generated by roxygen2: do not edit by hand
% Please edit documentation in R/create_model_utils.R
\name{load_cp}
\alias{load_cp}
\title{Load checkpoint}
\usage{
load_cp(
  cp_path,
  cp_filter = "last_ep",
  ep_index = NULL,
  compile = FALSE,
  learning_rate = 0.01,
  solver = "adam",
  re_compile = FALSE,
  loss = "categorical_crossentropy",
  add_custom_object = NULL,
  margin = 1,
  verbose = TRUE,
  mirrored_strategy = FALSE
)
}
\arguments{
\item{cp_path}{A directory containing checkpoints or a single checkpoint file.
If a directory, choose checkpoint based on \code{cp_filter} or \code{ep_index}.}

\item{cp_filter}{Condition to choose checkpoint if \code{cp_path} is a directory.
Either "acc" for best validation accuracy, "loss" for best validation loss or "last_ep" for last epoch.}

\item{ep_index}{Load checkpoint from specific epoch number. If not \code{NULL}, has priority over \code{cp_filter}.}

\item{compile}{Whether to load compiled model.}

\item{learning_rate}{Learning rate for optimizer.}

\item{solver}{Optimization method, options are \verb{"adam", "adagrad", "rmsprop"} or \code{"sgd"}.}

\item{re_compile}{Whether to compile model with parameters from \code{learning_rate},
\code{solver} and \code{loss}.}

\item{loss}{Loss function. Only used if model gets compiled.}

\item{add_custom_object}{Named list of custom objects.}

\item{margin}{Margin for contrastive loss, see \link{loss_cl}.}

\item{verbose}{Whether to print chosen checkpoint path.}

\item{mirrored_strategy}{Whether to use distributed mirrored strategy. If NULL, will use distributed mirrored strategy only if >1 GPU available.}
}
\value{
A keras model loaded from a checkpoint.
}
\description{
Load checkpoint from directory. Chooses best checkpoint based on some condition. Condition
can be best accuracy, best loss, last epoch number or a specified epoch number.
}
\examples{
\donttest{
library(keras)
model <- create_model_lstm_cnn(layer_lstm = 8)
checkpoint_folder <- tempfile()
dir.create(checkpoint_folder)
keras::save_model_hdf5(model, file.path(checkpoint_folder, 'Ep.007-val_loss11.07-val_acc0.6.hdf5'))
keras::save_model_hdf5(model, file.path(checkpoint_folder, 'Ep.019-val_loss8.74-val_acc0.7.hdf5'))
keras::save_model_hdf5(model, file.path(checkpoint_folder, 'Ep.025-val_loss0.03-val_acc0.8.hdf5'))
model <- load_cp(cp_path = checkpoint_folder, cp_filter = "last_ep")
}
}