Reorder training + testing recommendation steps

doc_source/deepar.md

@@ -54,10 +54,10 @@ Here, *q**i*,*t*\(τ\) is the τ\-quantile of the distribution that the model pr
 
 If you have a set of time series, as simple way to prepare, test, and train datasets is as follows:
 
-+ Use the full dataset in the test channel\.
-
 + In the train channel, remove the last `prediction_length` points from each time series\.
 
++ Use the full dataset in the test channel\.
+
 This ensures that the model does not see the removed points during training, and then those points are used for calculating the accuracy of the model\.
 
 For inference, DeepAR accepts JSON format with an `“instances”` field which includes one or more time series in JSON Lines format, and a name of `“configuration”`, which includes parameters for generating the forecast\. For details, see [DeepAR Request and Response Formats](deepar-in-formats.md)\.