I ran couple of simple ML examples available from the Documentation.
A small data set with less than 100 records would take a whole 10 sec to be trained in my laptop.
If I have a new set of sample data of 100 records available every hour, do I need to add those records to the population and train the entire Classifier all over again every hour? Is Classifier capable of retaining of its previous trained knowledge, while we keep feeding in newly available set of data to the Classifier every now & then?
On the site:
John McLoone explains the ML framework. In part 7 of the lecture at 3:25 minutes he shows that you can add data in a already trained network.
Hi I van Veen,
Thanks for your info. The given link seems broken. Been trying to click on it for the past few days.
However, after watching John McLoone's video, it seems like by adding one additional labelled data, he had to retrain the entire classification. My concern is the efficiency of retraining the classification. For instance, I have a sample model that takes about 3-4 hours to train. If I just want to add a few more sample data to the previously trained population, does it take equally the same amount of time to retrain if using this method of John Mcloone's?
Thanks I Van Veen,
I just tried using the same approach of John's. It works well with a much shorter time in retraining. Wonderful!