Training a CanonicalRNNEstimator model leads to the following error:
AttributeError: 'pandas._libs.tslibs.offsets.YearBegin' object has no attribute '_period_dtype_code'
Weirdly enough, using the same dataset to train a MQRNNEstimator model (with its extra estimator parameters) works fine. I have made sure that the pandas DataFrame is not irregular and that all time-steps for all item_ids do have non-NULL values. The only difference (apart from the parameters specific to each estimator) is that my version of MQRNNEstimator sets the seeds, whereas I am using the vanilla CanonicalRNNEstimator in GluonTS.
Description
Training a
CanonicalRNNEstimator
model leads to the following error:AttributeError: 'pandas._libs.tslibs.offsets.YearBegin' object has no attribute '_period_dtype_code'
Weirdly enough, using the same dataset to train a
MQRNNEstimator
model (with its extra estimator parameters) works fine. I have made sure that the pandas DataFrame is not irregular and that all time-steps for allitem_id
s do have non-NULL
values. The only difference (apart from the parameters specific to each estimator) is that my version ofMQRNNEstimator
sets the seeds, whereas I am using the vanillaCanonicalRNNEstimator
in GluonTS.To Reproduce
Error message or code output
Environment