Training on a large set of data would result in "Unable to predict" error.Issue Agent Intelligence is all about the data we train. The data should be of good quality and as a best practise, it is suggested to train atleast 50k records. Sometimes we end up training way too many records like 300k. The training might be successful but when you try to to test the results using REST API Explorer, you might get "Unable to Predict" error.CauseTraining too many records might cause the instance to run out of memory.ResolutionThere is no specific number to train the data but it is best suggested to have a good quality data with a minimum of 50k. Training too many records (< 200k records) would also fail to predict the result. If you feel you have trained a large data set, try to minimize the data and try training again.Related LinksRead KB0691436 to know the best practices.