Previous Next  Up  Table of Contents  SpamSieve Home

6.3.2   Auto-train with incoming mail

This option causes SpamSieve to automatically train itself with newly received messages based on their predicted categories. The whitelist will stay up-to-date so that messages from previous correspondents are never marked as spam. Also, after the initial training you would only need to train SpamSieve to correct its mistakes. It would automatically learn from the other new messages, as needed, trying to keep a proper balance of good and spam messages in the corpus. As the corpus increases in size, the rate of auto-training will automatically decrease to prevent it from growing too large.

Most SpamSieve users should keep auto-training on at all times. Reasons that you might turn it off include:

Testing Purposes
If you’re trying to figure out why SpamSieve is processing a certain kind of message in a certain way, turning off auto-training reduces the interactions among SpamSieve’s components and makes it easier to see what’s happening.
Unable to Correct Mistakes
When auto-training is enabled, it is imperative that you correct SpamSieve when it makes a mistake; otherwise it will “learn” things that aren’t true and begin making predictions based on that incorrect information. If your mail program and filters are set up such that you cannot tell when SpamSieve has made a mistake (or if you can see the mistakes but sometimes forget to correct them), then you should not use auto-training.
Multiple Copies of SpamSieve
If you are using multiple copies of SpamSieve (on different Macs) to filter the same IMAP account, it’s hard to know which one(s) to correct when a mistake is made. With auto-training off, you don’t have to worry about that.
Bulk Processing
For example, if SpamSieve will be processing a large number of incoming messages in a row, and you will not be able to correct any mistakes until later. In this case, using auto-training could magnify one or two mistakes into many, as SpamSieve would assume that it had classified the earlier messages correctly, and learned from that (possibly incorrect) assumption.
Delayed Corrections
If it will be a long time between when SpamSieve filters some messages and when you can correct any mistakes, the situation is similar to Bulk Processing, so you may want to turn off auto-training.
Previous Next  Up  Table of Contents  SpamSieve Home