Closed aamrb96 closed 3 years ago
I found a simple solution to this for everyone else that might be facing a similar problem. Don't read the checkpoint/trained model from the disk but simply run the loop with the trainer element from the first run again.
for data in files:
training = ...
validation = ...
train_dataloader = ...
val_dataloader = ...
trainer = ...
nets = NBEATs.from_dataset(training...)
trainer.fit(net, train_dataloader, val_dataloader)
Hi all,
I am currently working with the NBEATs implementation in this package. I wrapped all the training code from the tutorial into a function called `nbeats_train:
I would like to train it on three very large datasets (the datasets can not be appended into one large dataset due to hardware limitations).
Thus I wrote a loop that iterates over my data dictionary (
data = {"dataset_name" : TimeSeriesDataSet Object}
). After the first iteration, I would like to initialize the weights of the model after training it with each data set with the weights of the previous iteration. Thus, I wrote a loop like this:However, I receive the following error:
Does anyone know where the problem comes from/how to solve this?
Thanks in advance!