Manual Training of Bayesian Filter

Discussions on webmail and the Professional version.
Post Reply
chedesigns
Posts: 1
Joined: Wed Jan 20, 2021 2:08 am

Manual Training of Bayesian Filter

Post by chedesigns » Wed Jan 20, 2021 2:20 am

Hi:

This is no doubt a dumb question, but I thought I better clarify.

When you are manually training the Bayesian Filter (https://www.mailenable.com/documentation/10.0/professional/Manual%20training.html), is the training incremental?

By this, I mean which of the following scenarios is correct:

Scenario 1 :
On Monday manually train the Filter using 1,000 each of Spam and Ham messages. Delete the 1,000 messages used for training from your training folder
On Tuesday manually train the Filter using (say) 100 new/different Spam and Ham messages. Delete the 100 messages from your training folder
On Wednesday: repeat Tuesday, etc

So after the initial seeding of the Dictionary, each subsequent training is just with a new set of messages which builds up the dictionary incrementally

Or, Scenario 2:
On Monday manually train the Filter using 1,000 each of Spam and Ham messages. Leave the 1,000 messages in your training folder.
On Tuesday add a new (say) 100 Spam and Ham messages to the appropriate training folder. Now manually train the filter using 1,100 each of Spam and Ham messages. Leave all 1,100 messages in each training folder.
On Wednesday: repeat Tuesday ...

So, each manual training run is with a larger and larger set of messages. If you miss messages from a previous day's training, then the Dictionary loses the tokens from that previous training

I *think* it is Scenario 1, but if someone can give a simple confirmation it would be appreciated!

Post Reply