[Rspamd-Users] Questions about REDIS and Bayes Learning

[Alexander Moisseev]

On 19.11.18 3:02, Tim Harman via Users wrote:
> On 16/11/2018 9:43 pm, Manuel Garbin wrote:
> 
>> Hi Tim,
>> autolearn will autofeed the bayes DB, that mean that you have 200+ learned.
>> For your question :
>>
>> 1) I've used same command to clean a redis cache, it works
>> 2) disable autolearn if you want use manual learn
>> 3) use header + body and remove the X-spam status header
> 
> With regards to the first part:
> 
> 127.0.0.1:6379> HGETALL RS
> 1) "learns_spam"
> 2) "100"
> 3) "learns_ham"
> 4) "52"
> 5) "version"
> 6) "2"
> 
> This seems to imply that less than 200 of both have been learnt?  Or am
> I using the wrong REDIS command?
>

The command is correct. You can get the same counters using `rspamc stat` command.
At minimum, 200 ham and 200 spam mails need to be learnt.

> 
> With regards to 2, is it bad if I let it auto-learn but also feed it
> other emails (old ones) manually?
> 

You can use both auto- and manual- at the same time. Autolearn is bad by itself as it cannot detect false positives. If you have decided to use autolearn for some reason (e.g. you want to use rejected mails to learn), set high thresholds like "autolearn = [-20, 20];" to minimize FP impact.