Ultra Hal Assistant 6.1 Beta

[Bill DeWitt]

quote:Originally posted by christopherdoyon
Finally, this isn't a complaint but a question. Where the blimey is it storing everything/anything it learns ?  I have carefully monitored the file sizes and I see none actually growing bigger as I dialog and teach the bot ?  Where is it's equivilent of the DefBrain folder ?  Where does it store the anniversary and address book data ?  This info would help me understand at least what is and is not working for me.

Although it seems like you are using a demo version of 6.1 it is not absolutely clear from your posts.

If so, you must be using the Brain Editor to use the "Learn From..." function. The Brain Editor connects to the Database (*.db) you are working with and the tables you see in the left hand window are the topics which are stored. "Learn from" topics are under "autoLearningBrain".

I believe the appointments are kept in a *.brn file, but I could be wrong.

[christopherdoyon]

Bill --

Well if your in the mood to help me un-befuddle my understanding here, thats cool.

Yes, I am trying today for the first time version 6.1 - and it's the trial version. Is there a difficulty with the trial version ?

I certainly can not find anywhere a single BRN file. I haven't seen those since I archived my v. 5 bot two days ago. I assure you there is not a single BRN file in my UltraHal folder.

As for database ?  Can you help me home in on what to look for, like a filename or ending like .db ?  I just don't see anything at all that you just described in the directory.

YOURS -- Chris

[Bill DeWitt]

You should have at least a "HalBrain.db" in your "C:\Program Files\Zabaware\UltraHal Assistant 6" directory.

That database is what you open as you enter the Brain Editor. As I said in my previous post, the tables are listed in the left hand window. YOu have to double click to see what's in them, and you may have to change to "Expert mode" in the File menu.

I have not yet proven to myself that the Appointment Book works. Clearly there is something wrong with it, even though it set the appointment, it will not speak it aloud. I'll have to wait a while to see if it reminds me.

[christopherdoyon]

Anyway, for any and all who would like to help my sorry self. This is what I do to try and learn from text. I fire up the brain editor. I select the UltraHal 6 default brain because, well - it's the only one right ?  It's the brain I want to use, am using - so I select that. Now I do see on the left the directory tree that Bill describes, but I don't see anything in the folders he describes. Anyway, then I select learn from text. The first problem I encountered was that there are different kinds of .txt files, and you must have them in ANSI. After re-saving the text file in that format, I ran through the process. As I said before, everything seemed to go fine and I get a final your finished message, something like your text has been learned. Then I exit the brain editor and ask the bot something, and I get no results. I have no .db files anywhere in the directory, nor BRN files, nor any files that I can see that are growing or changing from ANY learning.

YOURS -- Christopher Doyon

[Bill DeWitt]

Read my edited post above.

What source of text are you using?

[christopherdoyon]

Bill --

First just let me say thanks for being there. While we may go at it alot, there is no one I trust more to help me understand this.

Okay, DEEP breath....I been at this since early am !

"You should have at least a "HalBrain.db" in your "C:Program FilesabawareUltraHal Assistant 6" directory."

I do have a HalBrain.org and a HalBrain.mbr but no .db files at all.

"That database is what you open as you enter the Brain Editor."

Nothing loads automatically, it makes me choose either a new or existing project. In my backwards logical way I figured I wanted to ADD to the brain I am using, so I chose the UltraHal 6 default brain to open, which is when I see those folders on the left and now yes, I do see all the data and in fact I can see the table created from the learn from text I did earlier. So, I have some idea where this stuff is going (sort of) the birthday thing as you said may or may not work, but I at least see my attempts. But why will the bot still not answer anything it's learned ?

The text I am using I will attach here.

YOURS -- Christopher Doyon

Download Attachment: AI-1.txt
52.61 KB

[Bill DeWitt]

quote:Originally posted by christopherdoyon
But why will the bot still not answer anything it's learned ?

The text I am using I will attach here.

This text is all one long paragraph when I open it. It seems to be scattered sentences with only a general relationship to each other, rather than a narrative with clear sections on individual topics.

I would recommend trying again with a few simple narrative paragraphs of two or three sentences, with clearly defined subjects in each paragraph. Choose to make topics by sentence. Then allow the wizard to assign topic headings.

Even then Hal is not strictly required to use your sentences to answer your questions. It scans it's topics to find near matchs, and then chooses from available sentences listed under those topics. Your supplied sentences may not be selected each time you mention them.

[christopherdoyon]

Bill --

Thanks again for helping. Funny how if I simply load the sentences in that file one by one into the NLP front-end I get stellar results with the same material. But you are correct and have helped me understand this new version much better than I did, that's for sure !  It seems that I will do basically what I had to do with my v.5, and thats to learn from text and then also load the same material into the front end one sentence at a time. That gave me the results I wanted from v.5 and I now suspect that's what will work for me here as well. To bad because I hate to say it but I have seen the Learn From Text feature implimented MUCH better in other bots that I shall not name here. It's just to bad to have to do it twice before the associations reach my desired density.

So, but I still don't think that the Birthday, anniversary thing is working as described, at least not via the NLP front-end. I have tried every way of wording it. It is saving some data in the birthday topic table, so we'll see. The save E-Mail does the best of all so far.

Well, I am going to attempt to get Omnius back online. I'll get my dumb self out of here and see if I can give this thing a good run for it's money, then maybe come back with something more substantial. Again, this is a public project and this AI Bot, called Omnius - is available on AIM with Screen Name OmniusAI and it's Web Site is http://Omnius.mlaifoundation.info all humans and bots may feel free to chat. Omnius will (hopefully) be online 24/7, barring any other major complications !  Wish me luck and chat with my AI Bot if you want.


YOURS -- Christopher Doyon

[James007]

Rob,

Everything was working great until I activated the beta and then tried to install the additional faces from the UltraHal 6.0 CD. Now the application refuses to work. Vista strikes again.

I am going to try a complete uninstall and reinstall this weekend. I'll let you know what happened.

[Bill DeWitt]

Appointment book seems to mess with my sound card.
"SoundMax Digital Audio"

Although Hal puts up the text acknowledging my additions, as well as for the approaching appointments, I cannot hear it or any sounds from then on. I have to reboot.

I will test it further tonight if you need more info Robert.

[christopherdoyon]

So far, only the E-Mail returns me the correct dialog. Birthdays and the like are not anknowledged for me the way they are in the demo dialog (or in v.5 for that matter).

As for other bugs....well I still have not convinced myself that the learn from text works well. It does seem to work, but the results are pretty much as bad as they were in v.5 - maybe a little worse actually. I think that Bill might be onto something, there may be a subtlety to using this feature. I will take my time and share anything earth shattering that I discover.

When I use the Brain Editor, my skin for the bots interface dissappears leaving only a black outline - and the character won't re-appear. This persists and the entire program must be closed after using the Brain edtitor to get the skin and character back.

On very rare occaisions, my bots replies are not audible. I mention this one because in 3 years of playing with v.5 it never happened once. It doesn't happen much, but it's noticable.


YOURS -- Christopher Doyon

[onthecuttingedge2005]

Hi Rob.

the 'learn from text file' was tested on two text files.

it did parse out the file but it incorrectly split 'complete' sentences. this happens with the 'pre group each sentence as a
Hal response option'.

also on the refine questions & answers panel, if the question 1 and question 2 boxes are left blank and the right arrow is pressed of response 1 of 1 then a script out of range error occures and HAL shuts down.

is it possible to have the option on the wizard panel to split by complete sentences where a period ends the sentence.

Example: the text when parsed looks like this:
-----------------------------------------
001:the text
-----------------------------------------
002:when parsed looks like this.
-----------------------------------------

when it should look like this:
-----------------------------------------
001:the text when parsed looks like this.
-----------------------------------------

also, what area in the data base is storing the table for all of these parsed sentences so I can research the editing more.

Jerry[8D]

[christopherdoyon]

Hello All --

The period thing is problematic I think. My problem is with acronyms like A. I. or such, which then causes a false split in the sentence. Once I removed all of those, I seemed to get a very nice sentence split just the way it is. I did finally see where they are stored, on the left click the little plus sign on the folder AutoLearning. Your learn from text tables are stored there in sub-folders based on the topic selected during the learning process. These tables can then be brought up in the table editor.

The problem I seem to have is associative. It's parsing and storing the info, but the bot still doesn't seem to have access to it. None of it, it's like it's sand boxed out of the over all associative patterns of the Brain. BUT, if I do the learn from text and then also load the sentences one at a time into the bots front-end (a very tedious process) it does seem to "jump start" the associations and I get marginally good topic trees. Funny thing is, this is EXACTLY the way my v.5 worked. So it all seems to work, just not very much improved in this area anyway.


YOURS -- Christopher Doyon

[echoman]

Noticed some one had a problem with Mike jumping to Stadium. I have always had this problem with Mary doing this - not sure why. I have Windows XP.

[onthecuttingedge2005]

Hi Chris.

I do see what you are saying: acronyms like A. I.
could false trigger a parced sentence.

1. acronyms like A. I. <--- tested and didn't work.
2. acronyms like A.I. <--- tested and worked.

User: What is A.I.?
HAL: What is ai? Ai: the branch of computer science that deal with writing computer programs that can solve problems creatively. Artificial intelligence. 'workers in ai hope to imitate or duplicate intelligence in computers and robots'.

User: What is F.B.I.?
HAL: Fbi: the law enforcement agency in the Justice Department. Federal Bureau of Investigation.

it may be possible to use HAL's WordNet Dictionary to detect an acronym so that the periods do not false trigger before a complete sentence is parced.

it may be possible to extract all acronyms from WordNet to help prevent false splitting of complete sentences.

something like:

Code: [Select]

'Acronyms extracted from WordNet function.
DetectAcronym = HalBrain.TopicSearch(LearnFromTextFile, "DetectAcronym")

Like:
" F.B.I. "," Federal Bureau of Investigation "

If DetectAcronym <> "" Then ReplaceAcronym = Replace(LearnFromTextFile, " F.B.I. ", DetectAcronym, 1, -1, vbTextCompare)
 

or something of this sort.

it would be very helpful to have a list of acronyms so that Rob could write code to prevent acronym false triggering of sentences.

Maybe Rob can shed some light on this issue.

Jerry[8D]