This message was deleted.
# neural-forecast
s
This message was deleted.
a
And I have a complementary question concerning the DeepAR implementation. I'm having trouble understanding the difference between the
trajectory_samples
parameter and the
num_samples
that can be set in DistributionLoss 😅 Thank you in advance!
c
Hi @Antoine SCHWARTZ -CROIX- 🙂 . The library does not offer currently a way of accessing the samples directly. You could use the
sample
method of the
DistributionLoss
class by passing the distribution parameters, but it might not be straightforward to get the correct inputs. This is a great feature so we will add it in the future. Can you make an issue with this request please?
a
Thanks for your answer @Cristian (Nixtla), do you know a way to get these
distr_args
in the correct format for calling the
sample
method?
(I've just opened the issue)