Attrasoft Discussion Group I

Thank you for visiting Attrasoft Discussion Group.
To add your comments, click here!
To read messages, please continue:


Dear Dennis:
 
Thank you very much for your feedback. I wish you good luck in the future.
 
>>I feel that how the data is laid out must have a bearing on what the analysis finds.
 
I agree with you that predict 1 column at a time might not be a good choice. To predict several columns at a time, the Predictor requires more data than a single column. For example, let us say we want to predict 6 numbers at a time, each number has 50 different values, we can estimate what the number of rows required will be:
 
10 * 6 * 50 = 3000 rows.
 
If you do not have enough data, then you will not get any results from the Predictor. You have two choices, (1) Predict 1, or 2, or 3 at a time instead of 6 at a time; (2) reduced the number of values from 50 to a smaller number. In your case, you can reduce 49 to 5 numbers: (1-9,10-19,20-19,30-39,40-49), which is the second choice.
 
There is a third way, which I have used several times. This combines the Predictor with the DecisionMaker. This will work when the Predictor can only pick up 3 or 4 numbers; I will then use the DecisionMaker to pick the remaining numbers.
 
Good luck.
 
Gina
 
 
----- Original Message -----
From: DJT
To: webmaster@attrasoft.com
Sent: Friday, October 11, 2002 7:41 PM
Subject: Predictor 2.6
 

I have the standard version of 2.6 and have been using it with our Pick3 game. I have done pretty good so far.
I want to apply it to our 649 game. I personally don't think it will be effective in analyzing the column by column numbers of the lottery draw since there is such a swing in the numbers. . What I have done is written a program to put the numbers into decades (1-9,10-19,20-19,30-39,40-49) and then run the Integer analysis on that data. I normally get 3 out of 15 and have had as much as 5 using this method. But, I know little of neural networks other than they are a "pattern" analysis. The biggest trend I have been able to use is 7 but the results are normally more consistent at 5. I am thinking about breaking the data down into sections of 7 instead of 10. I don't know if there is any validity in what I am doing and I would ask your opinion on it. I feel that how the data is laid out must have a bearing on what the analysis finds.
I have not really had much success in using the + and - method and looking for the most likely smallest and biggest pick. It seems to do better if I just use the best two choices from each decade.
 
Thanks for the great program at such a reasonable price, that is so easy to use and well-documented.
As my winnings increase I intend to purchase the next version up.
 
Dennis Tretiak
Nipawin, SK. Canada
 



Answer:

My suggestion is rollback: Delete the last 6 numbers; save and run. See if
the file works. If it does, then re-enter the last row of numbers.
Otherwise, you might have accidentally deleted some number in the middle,
that will take more time to figure out.

Another possibility: how many rows of data do you have? the number of rows
in the standard version should be less than 10,000. I do not know if this is
your problem.

Have a Nice Day,

Gina

Question:
> Dear Sir :
> I have purchased Predictor 2.6 some time ago and I use it to play Lotto.
> I'm happy to tell you that the Predictor's predictons are very accurate.
> But I have a problem. As you know, every week I must update my Lotto file
> witha new draw. Well, when I add the last draw in the file, I receive a remark
> thatsays : "last row is not completed", or something like that. And "Please see
> Initialize/test Input File". But the truth is that the row that I entered is complete.

> I entered the 6 numbers. So, I don't know why I receive such a remark ! And
> of course, Predictor don't give me any predicton under such conditions.
> Please, instruct me about that problem. What can I do to update correctly my
> draws' file ?
> Thank you very much


Answer:

Software: Attrasoft Predictor 2.6.

Symptom: run an example by clicking “Example/ …”, and get an error "File I/O Error file not loaded".

Reason: All examples are stored in the default directory, C:\Program Files\Attrasoft\Predictor 2.6\. If you leave this directory by clicking “File/Open” or “File/Save As”, you will get this error.

Solution: Close the program and restart (you will be back to the default directory), you will be able to run the examples.

Q:I followed your instruction and it worked with the sample data that was
given to me on the disk.  However when I am using my own data, I open it,
link it,and test it and then click "example markov chain" and I get an error
"File I/O Error file not loaded" and in the title bar it has "
D_math2a.txt". How do I get it to recognize my files?
thank you


Dear Gina,

Special thanks for your advise and prompt help.

I realy apreciate your free help and ofcourse I will mention Attrasoft, Gina

in my desertation under "ACKNOWLEDGEMENTS"

I'am very happy with these PREDICTED results "within 2"

This is PERFECT !

Ofcourse to handel/run my large datafile I will need the updated Predictor 100K Version.

Can you give me a special (non profit) price for the 100K Version ?

Thanks

Best regards from Holland/Europe


You stated:

“This process is repeated again and again to see how mother nature is fluctuating

the growth. In manual charting I found remarkabel patterns in the historical data

and I believe in case there is a certain growth-pattern it is predictibel.”

For better prediction, we want to eliminate the fluctuation.

To do this, I use a 5-average. Please see attached data file.

I isolated the last 10 numbers for testing.

Parameters:

Precision level = 36

Trend = 15

It follows when the numbers are increasing, one should use “-prediction”.

When the numbers are decreasing, one should use “+prediction”.

The commands are "10-average/+Integer" and "10-average/-Integer".

When the numbers are decreasing, "10-average/+Integer" will be used. When the prediction curve turnsflat, stop using this command and change to "10-average/-Integer".

The last 10 integers are:

20

17

15

13

12

19

21

23

25

25

The process will take two runs. The first 5 are decreasing, so the command is "10-average/+Integer".

The results are:

18668442

16594282

14389114

12288664

12176986

12175636ç== flat, discard

12121332

12118708

1195084

1169248

Append the following to the data file.

20

17

15

13

12

Now the next 5 are increasing, so the command is "10-average/-Integer".

The results are:

17118888

20142932

21179984

23130084

2383380

2363480ç== flat, discard

2340254

2347012

2344452

24216392

Together:

RealPredicted

2018

1716

1514

1312

1212

1917

21           20

2321

2523

2523

All predictions are within 2.

This result is about as good as it gets.

If you run 10 times instead of two times, you might be able to get a better result. It is hard to beat “within 2”.

You have one alternative: since your computer collects one reading every 10 seconds, it is possible for you to collect 250,000 measurements, even 1,000,000;you might getbetter results with a larger data file.


I will run the 100K Predictor soon.

A:

I'am very curious for these results.

 I have 2 questions:
 

 

(1) When you collect 37 measurements, you mean you measure the same variable 37 times, say once every 39 minutes around the clock, correct?

A:

YES, the same variable 37 times, once every 60 seconds around the clock.

This is a full automatic process.

What I need to know is: you are not reading from 37 different instrument readings, are you? but rather reading the same instrument repeatedly, correct?

A:

YES, I'am reading the same instrument repeatedly. 

Background forquestion 2.
 

 

It is important to collect all variables. Otherwise, the prediction will fail. Here is an example in the Predictor:

A:

I understand this is very important.

Question 2: why will the historical values of the growth rate determine the future growth rate? Why are no other factors involved in the growth rate? Or why are all other factors negligible?
A:

My theory:

I believe that the growt rate has a limited permutation and this is NOT a random process.

I repeat again and again this process to register the outcome and see how mother nature is fluctuating 

Finally, I hope to find the basic patterns in my data.

Other Factors? Possibel parameters could be temperature and atmospheric pressure.

These 2 parameters are constant in a clean room under laboratorium conditions so we can 

neglect this.

I hope this will help.

Best regards, Sandor Kovacs
 

 

I will run the 100K Predictor soon.

I have 2 questions:

(1) When you collect 37 measurements, you mean you measure the same variable 37 times, say once every 39 minutes around the clock, correct? What I need to know is: you are not reading from 37 different instrument readings, are you? but rather reading the same instrument repeatedly, correct?

Background forquestion 2.

It is important to collect all variables. Otherwise, the prediction will fail. Here is an example in the Predictor:

Let a system be defined by a triplet ( x, y, z ). They obey:

x (n+1) = ( x (n) + y(n) + z(n)) Mod 7;

y (n+1) =x ( n );

z(n+1 ) = y ( n) Mod 5.

Starting from (1 0 0), here is the generated sequence:

*

*

3

100

110

211

421

042

604

360

231

623

462

541

354

530

153

210

321

632

463

641

464

041

504

250

020

202

420

642

564

151

010

101

210

321

632

463

641

464

041

504

250

020

202

420

642

564

151

010

101

210

321

632

463

641

464

041

504

250

020

202

420

642

564

151

010

101

210

321

632

463

64 1

Run the Predictor and click “10 Average/+ Linear”, we obtain:

464384000

041384000

504384000

250384000

020384000

202384000

420384000

642384000

564384000

151384000

This result is 100% CORRECT.Now let us use an INCOMPLETE model, say, y-column only. Here is the data:

*

*

1

10

1

1

2

4

0

6

3

2

6

4

5

3

5

1

2

3

6

4

6

4

0

5

2

0

2

4

6

5

1

0

1

2

3

6

4

6

4

0

5

2

0

2

4

6

5

1

0

1

2

3

6

4

6

4

0

5

2

0

2

4

6

5

1

0

1

2

3

6

4

Now run the Predictor and we get:

625120

424832

124674

524576

224672

124576

224576

424576

530720

41280

Compare this with the correct answer:


 
6
4
0
5
2
0
2
4
6
5

We can see the problem.

Question 2: why will the historical values of the growth rate determine the future growth rate? Why are no other factors involved in the growth rate? Or why are all other factors negligible?


in the range between 0-36, so I have 37 variabels." 

If your data is enzyme growth rate in digital form 

in the range between 0-36, then you have ONE variable with 37 values, not 37 variables. Please explain, ONE variable or 37? 

A:

You are absolutly right.....

I have ONE VARIABLE with 37 VALUES.

Sorry for my mis-interpretation.

If one variable, 

why the historical values of the growth rate will determine the future growth rate?

A:

I have isolated enzyme and injected with a chem.solution

and now I start to  analyse/observe  read the growt-rate represented

in digital form from value between 0-36.

After I have collected 37 measurments (this is one round/cycle)

I reset/callibrate and restart repeat the process for the next round/cycle to observe

again 37 measurments to see how the growt-rate develops.

This process is repeated again and again to see how mother nature is fluctuating

the growth.

In manual charting I found remarkabel patterns in the historical data

and I believe in case there is a certain growth-pattern it is predictibel.

If you can, please read the manual for the prediction theory part, so we can communicate better.

A:

Thanks and I will take a closer look.

I hope this information will help.

I checked your data file with the assumption that the data is the "growth rate", and the data model has one and only one variable. 

A:

That's correct.
 

 

The data file contains about 10,600 numbers. 

A:

Yes and I have used this dataset with Predictor Vers. 2.6 
 

 

Can you send us a data file which has all 37,000 numbers.

A:

Yes,  I will send it as attachment my coplete dataset of 37000 measurments.
 

 

Also, the Predictor 2.6 has 10,000 limits on several places, the number of rows in the data file must be less than 10,000. 

A:

I did NOT know this limit.

In the manual I read that Predictor can handel terrabyte of data.
 

 

I will use the 100K version to run your data on 37,000 integers.

A:

I'am very curious for these results.
 

 

On your part, please justify WHY the historical growth rate can determine the furture growth rate??????

If you cannot justify this, your approach has a problem. Please read the Predictor manual for the prediction model.

A:

Read my explanation in before last e-mail.


Question: you said "My data is representing enzyme growth rate in digital form

in the range between 0-36, so I have 37 variabels." 

If your data is enzyme growth rate in digital form in the range between 0-36, then you have ONE variable with 37 values, not 37 variables. Please explain, ONE variable or 37? 

If 37 variables, what are they? 

If one variable, why will the historical values of the growth rate determine the future growth rate?

I checked your data file with the assumption that the data is the "growth rate", and the data model has one and only one variable. The data file contains about 10,600 numbers. Can you send us a data file which has all 37,000 numbers.
 

 

Also, the Predictor 2.6 has a 10,000 limit: the number of rows in the data file must be less than 10,000. I will use the 100K version to run your data on 37,000 rows.
 

 

On your part, please justify WHY the historical growth rate can determine the future growth rate??????

If you cannot justify this, your approach has a problem. 
 

 

I'am using Attrasoft Predictor Vers.2.6 and need your help

 to set-up Predictor Vers. 2.6 for the best forecast.

Enzyme Research:

My data is representing enzyme growth rate in digital form

in the range between 0-36, so I have 37 variabels.

I have already collected 37.000 measurments.

Details on my data procedure see below in my before last message.
 

 

See my collected data as attachment file: Enzyme Data1.txt

Continue my Process:

Datafile will be updated after I have collected 37 NEW measurments.
 

 

For my research I will need the

BEST POSSIBEL PREDICTION for the number(s) in the next row.
 

 

Please advise HOW to achive this.

Is my data formatting/ Setup/handling correct?
 

 
 
 
 
 

*

Enzyme Data1

*

1

19

18

34

31

36

36


 Thanks
 Best regards
 


Erik....What application did you use the DecisionMaker for Doc?.........Britney is great!

Wole....Are you using the DecisionMaker for Stocks or lottery?

Wiemel...You said the DecisionMaker was useful. What is your application?
Gina <gina@attrasoft.com>
- Wednesday, June 28, 2000 at 18:54:25 (EDT)


I was recommended DecisionMaker by a friend of mine, and I have to admit: on the whole it definitely is a fine and useful program. Still, I found an annoying bug this morning. I was still somewhat sleepy and had a terrible hangover. I didn't want to go to work, so I asked DM "Should I stay or should I go?". DM decided I'd better go to work. Any idea how to solve this problem? Thanks. BTW. Did you know thery're building a Britney Spears museum in her hometown?
Wiemel
Zweten - Wednesday, June 28, 2000 at 12:51:40 (EDT)

You're right, Erik! But I guess Brittney is more of a Microsoft gal!!! (hahahaha!) I use DM almost every day and I would love to learn more about the other program you wrote about! I would love to learn more about the other program you wrote about! Oops I wrote it again :)
Wole
Sierra Leone - Wednesday, June 28, 2000 at 08:18:38 (EDT)

Loved DecisionMaker! The name of the program reminded me of another well known software product: D-cide4U(www.D-cide4U.be). I'll encourage all my friends -who are also doctors- to use your product. Do you like Britney Spears?
erik </>
leuven, belgium - Monday, June 26, 2000 at 08:23:58 (EDT)

This is a very nice place !
Protector 1uno <spavle@nightmail.com>
Gospic, Li Croatia - Monday, May 22, 2000 at 00:37:33 (EDT)

Dean:

Possible reasons:

(1) You are using NT or Windows 2000: a NT version has to be ordered;

(2) If you are using Win95/98, but win95/98 is not installed in C drive: create a folder C:\Windows\Temp\.

This should solve your problems.

Ying
Ying <ying@attrasoft.com>
Savannah, GA USA - Wednesday, May 17, 2000 at 12:49:29 (EDT)


I am using ImageFinder 4.4 and like it very much but for some reason it won't display results in a web page. It just displays in the small window in lower right and says: IO Exception. Any Ideas?
Dean <ktbaby@telus.net>
USA - Tuesday, May 16, 2000 at 04:53:25 (EDT)

Ali:

Buildings, roads, rivers, ... Yes.

Cars: 1-meter resolution might not be good enough.

Ying
Ying <ying@attrasoft.com>
Savannah, GA USA - Wednesday, May 16, 2000 at 12:49:29 (EDT)


We are looking for developing a software for image recognition of objects from a Satellite imagery. Lets say we have got a 100 x 100 Km, 1-meterresolution file (in TIFF format). The file may be 128 MB in size. We would like the software to identify all the { buildings, roads, rivers, cars, and other items }from the satellite image. What is your openion, how is it possible, can you help us develop it.
Ali nasir <ansr@tx.micro.net.pk>
wah cantt, Turkey - Friday, May 12, 2000 at 13:45:35 (EDT)

Attrasoft ImageFinder 4.4 works for all images, including bank notes. If I can help, please email me. Thanks.
Gina <gina@attrasoft.com>
Savannah, GA USA - Thursday, March 02, 2000 at 09:13:37 (EST)

im interest in banknote recognition using inductive learning. if you have information about this topic please contact me...
Mehmet Sevkli <msevkli@fatih.edu.tr>
istanbul, turkey - Wednesday, March 01, 2000 at 07:30:13 (EST)

The new Dow5 is available from, see http://attrasoft.com.

Gina <gina@attrasoft.com>
Savannah, GA USA - Thursday, March 01, 2000 at 07:13:37 (EST)


The Dow5 seems to be very interesting and profitable:) But the approach is not very flexible. The price of 'cat' in this 6 months period drops quit a lot. Is it better to perform revision using the NN software more regularly? Will you continue to do it?
Richard Wilson <wilsonr@hotmail.com>
USA - Thursday, February 03, 2000 at 13:44:48 (EST)

Hi, Daniela L. De Luise, the short answer is yes.
Gina <gina@attrasoft.com>
savannah, GA USA - Monday, November 08, 1999 at 12:11:48 (EST)

Do you know if it is possible to solve this problem using NN? There is a round platform with many people getting on and down constantly. Which is the quantity of persons at time point 't' ? The problem must be solved using a video camera and a frame grabber.
Daniela L. De Luise <daniela_ldl@hotmail.com>
Buenos Aires, DC Argentina - Sunday, October 31, 1999 at 19:37:59 (EST)

Content-based Retrieval

Operating an imagebase is very similar to operating a relational database. Therefore the administration of an imagebase is very similar to the administration of a relational database. ImageFinder adds "Content-based image retrieval" (content-based query) to a relational database. A relational database already has "Keyword based retrieval" (query) and image management abilities. For info on the administrative end of a relational database, many books are available, say, Microsoft Access. Gina
Gina <gina@attrasoft.com>
Savannah, GA USA - Tuesday, July 13, 1999 at 12:20:42 (EDT)


I am interested in determining if ImageFinder will meet my needs for a photo library. Is there any info on the administrative end?
Megan Shisler <mshisler@nmpinc.com>
Falls Church, VA USA - Monday, July 12, 1999 at 15:34:01 (EDT)

>>I was looking at the WBCD since it seems to me there are enough examples for a 70:20:10 split of train:test:validate in that database. Any comments. << We split into train:validate only, where train = "train:test". WBCD = Wisconsin Breast Cancer Database. Thanks. Gina
Gina <webmaster@attrasoft.com>
savannah, GA USA - Thursday, February 25, 1999 at 15:26:05 (EST)

I like your website for its information. The examples are interesting but normally one would split the larger databases three ways to see if the net would actually predict anything with a validate set. Have you done this and what were the results. I was looking at the WBCD since it seems to me there are enough examples for a 70:20:10 split of train:test:validate in that database. Any comments.
Marc Parham <mparham@tiac.net>
Bedford, MA USA - Tuesday, February 23, 1999 at 09:15:02 (EST)

Hello ! I would like to meet someone having used Decision maker for horse races. Regards, A; Proux
A. Proux <aprx@club-internet.fr>
Paris, USA - Saturday, November 14, 1998 at 05:35:34 (EST)

Please add your comments to this discussion group. Thanks.