load_cp: Load checkpoint

In GenomeNet/deepG: Deep Learning for Genome Sequence Data

Load 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.

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

cp_path	A directory containing checkpoints or a single checkpoint file. If a directory, choose checkpoint based on cp_filter or ep_index.
cp_filter	Condition to choose checkpoint if cp_path is a directory. Either "acc" for best validation accuracy, "loss" for best validation loss or "last_ep" for last epoch.
ep_index	Load checkpoint from specific epoch number. If not NULL, has priority over cp_filter.
compile	Whether to load compiled model.
learning_rate	Learning rate for optimizer.
solver	Optimization method, options are ⁠"adam", "adagrad", "rmsprop"⁠ or "sgd".
re_compile	Whether to compile model with parameters from learning_rate, solver and loss.
loss	Loss function. Only used if model gets compiled.
add_custom_object	Named list of custom objects.
margin	Margin for contrastive loss, see loss_cl.
verbose	Whether to print chosen checkpoint path.
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.

Examples

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")

GenomeNet/deepG documentation built on Dec. 24, 2024, 12:11 p.m.