SiD
November 30, 2020, 12:58am
1
I could save weights in each epoch during training using keras R. I have attached code for callback_model_checkpoints() and fit() -
# callbacks to save weights in each epoch
callback_model_checkpoint(
filepath = "weights.{epoch:02d}-{val_acc:.2f}.hdf5",
monitor = "val_loss",
verbose = 0,
save_best_only = TRUE,
save_weights_only = TRUE,
mode = c("auto", "min", "max"),
period = NULL,
save_freq = "epoch"
)
# fitting the model on the training dataset
model %>%
fit(train_x, train_y,
epochs = 150, batch_size = 1024,
validation_split = 0.2, verbose = 0,
callbacks = list(callback_model_checkpoint("checkpoints.h5")))
But how do I load the weights for further diagnosis? I can see a file 'checkpoints.h5' in the working folder.
keras::load_model_weights_hdf5()
or keras::load_model_hdf5()
, depending on whether save_weights_only
is TRUE
or FALSE
in callback_model_checkpoint()
, respectively.
SiD
November 30, 2020, 3:06pm
3
thanks for replying @mattwarkentin
But I am getting errors. I am in the same working directory and there exist a file with name 'checkpoints.h5' -
keras::load_model_weights_hdf5("checkpoints.h5")
Error in normalize_path(filepath) :
argument "filepath" is missing, with no default
keras::load_model_hdf5("checkpoints.h5")
Error in py_call_impl(callable, dots$args, dots$keywords) :
AttributeError: 'str' object has no attribute 'decode'
Can you try load_model_hdf5("checkpoints.h5", compile = FALSE)
?
SiD
November 30, 2020, 4:32pm
5
hi, I am getting error
load_model_weights_hdf5("checkpoints.h5", compile = FALSE)
Error in load_model_weights_hdf5("checkpoints.h5", compile = FALSE) :
unused argument (compile = FALSE)
load_model_hdf5("checkpoints.h5", compile = FALSE)
Error in py_call_impl(callable, dots$args, dots$keywords) :
AttributeError: 'str' object has no attribute 'decode'
system
Closed
December 21, 2020, 4:32pm
6
This topic was automatically closed 21 days after the last reply. New replies are no longer allowed. If you have a query related to it or one of the replies, start a new topic and refer back with a link.