Transformation result keys names conflict about gluonts HOT 6 CLOSED

It appears that your dataset has no column with the name target, which is required.

from gluonts.

Thank you for reply jaheba,
It's obvious that my data don't have the name target.

Before applying the transformation my data have target but after applying the transformation it's getting splitted into past_target and future target which corresponds to context and prediction length

from gluonts.

@hari7696 when you call estimator.train, the estimator-specific transformation is applied internally: there is no need to apply the transformation by hand.

What happens in your example is that you apply a transformation outside, the "target" field gets split into "past_target"/"future_target" and disappears, and this transformed data is once again passed through a transformation inside the call to estimator.train, which expects "target" to be there.

from gluonts.

@lostella
Hi Lostella
Thanks for the reply
This is the estimator I am using
estimator = DeepAREstimator(freq="4H",context_length = 48, prediction_length=24, distr_output = gluonts.distribution.GaussianOutput(), trainer=Trainer(epochs=50))
As you said the estimator specific tranformation will be applied to the data.
But I cant find any arguments except time_features related to transformer in DeepAREstimator
refered doc link : http://gluon-ts-staging.s3-accelerate.dualstack.amazonaws.com/PR-93/1/api/gluonts/gluonts.model.deepar.html

can you give me some info on customizing the estimator specific transformer, I would like to add observed_features, age, is_pad and time_features to my data.

Thanks

from gluonts.

@hari7696 those features are automatically added by the transformation. In your example above, just do

predictor = estimator.train(train_ds)

without applying any transformation by hand to your data.

We will look into improving the documentation on how to correctly use estimators.

from gluonts.

Got a picture after looking into source code of DeepAREstimator().

Thank you

from gluonts.