[pandorabots-general] Generic Reductions for who and what... so far

[R. B. Chandler]

It is nice to know what I had in mind will work.  It tested out ok too.  
That knowledge would have been really helpful Sat. while I was waiting 4 
hours for my car to get fixed.  Ah, the problems of learning something new.  
My poor dictionary has been through a lot of changes encluding being 
completely scraped and started over.  Oh well, Rome wasn't built in a day.

Actually, I went with DEFINE ISAAC NEWTON to reduce all the WHO IS *, WHO IS 
* *, and WHO IS * * * so not to confuse Juice or Wayne with Isaac should 
(when) I create more topical dictionaries for my bot.  Any other variations 
on a given name can be added later independant of the generic data set.  
While my bot is already "live" I'll get more into trageted training once I 
get more of the basic framework done for my bot.  At this point there is no 
point in training it without the basic software in place and working.

I should finish this project by the end of the week if I don't go blind 
staring at this screen first.  Then I go on to learn how to make my bot 
control two other Iframes from its own Iframe.  After that it is crew and 
ship information from the story of which my bot is a part.  Then then I take 
on the current data sets (only the ones I need) converting them to English 
only (I'm surprised you didn't separate languages to begin with) and 
expanding each set at the same time, if needed, to suit the design of my web 
site and the story being told.  Once I'm done, my bot's general data sets 
will probably be more suited to businesses or web sites a little more on the 
serious side.  On the other side I doubt that I will ever be truly done as 
there will aways be more knowledge I can add.

Everyone's suggestions have really helped the learning curve a lot.  Thank 
you one and all.  I hope the running documentation of my project I've been 
doing has been helpful to others too.




>From: "Dr. Rich Wallace" <drwallace at alicebot.org>

>
>It looks like you are doing some good work.  At some point you may want to
>try the Targeting approach after you launch your bot.  Your clients or
>customers may drive your choices about what kinds of new patterns to add,
>and in what order.
>
>For example, you may find that that the category
>
>WHO IS *
>is activated
>
>fairly often, and you might also find out that the inputs include a lot
>of examples like
>
>WHO IS SIR ISAAC NEWTON
>WHO IS NEWTON
>WHO IS ISAAC NEWTON
>
>so your categories like
>
><pattern>WHO IS * *</pattern>
>and
><pattern>WHO IS * * *</pattern>
>
>might reduce those to
>DEFINE NEWTON
>
>using <star index="3"/> and <star index="2"/> but you might run into
>problems with inputs like
>WHO IS THE FAMOUS SIR ISAAC NEWTON
>
>It might be safer to write specific categories for the variations you
>actually encounter.  One rule of thumb of botmastering, sometimes called
>"continuous beta testing", is that it's hard to predict in advance what
>patterns you'll need to capture the inputs people will actually say to
>your bot.  Targeting is a way to capture these inputs once your bot goes
>live.