Cancel
Showing results for 
Search instead for 
Did you mean: 
Highlighted

Data fit tool issue

Valued Contributor
Valued Contributor

Dear community members,

I'm trying to use the data fit tool to estimate the distribution of interarrival times and I'm getting a problem:

when I go in the evaluation tab of tha distribution fitting tool, it always gives me the results for following distributions

 

Gumbel, Logistic, Cauchy, Laplace, Normal, Triangle, Uniform

 

But I want the results also of other distributions (beta, weibull, gamma). I already tried to select them in the fit tab -> distribution -> select, but when I click on "apply" it is like nothing happens and it gives me always the above distributions in the evaluation tab.

I'm I doing any mistake?

 

Thank you

8 REPLIES 8

Re: Data fit tool issue

Gears Esteemed Contributor Gears Esteemed Contributor
Gears Esteemed Contributor

if you don't see a distribution, then your data doesn't fit for it; you can have a look on your data, perhaps you have to delete some "outperformer" or failures

Steffen Bangsow
freelance simulation specialist  
web: www.bangsow.eu
mail: steffen@bangsow.net

Re: Data fit tool issue

Valued Contributor
Valued Contributor

I used the filter and now it gives me these output:

Cattura.PNG

The strange thing is that with another program I obtained as best fit a beta distribution multiplied by another variable.

Anyway, if I want to use the results in this table I should use the Weibull right? If so, how should I write the parameters in the source?

 

Thank you 

 

Re: Data fit tool issue

Gears Esteemed Contributor Gears Esteemed Contributor
Gears Esteemed Contributor

You should use Gumbel, it is the best fitting distribution, but you can also use Weilbull.

In the source, field interval select Gumbel and enter the parameters from the columns 13 and 14.

If you select the distribution, it is shown the order of parameters, you have to enter.

Steffen Bangsow
freelance simulation specialist  
web: www.bangsow.eu
mail: steffen@bangsow.net

Re: Data fit tool issue

Creator
Creator

Hi Mr. @Steffen_B  

 

I have a similar problem than Alessandro when I estimate the best distribution for my data these are the possible distributions.distribution.png

 

Then Do I need to use the LogNormal distribution for my data?  

 

I have this doubt, in any of two cases Can I choose the LogNormal Distribution as the best fit?

 

What does it mean the true or false result in each case?

 

Regards

Antonio

 

 

Re: Data fit tool issue

Gears Esteemed Contributor Gears Esteemed Contributor
Gears Esteemed Contributor

AD is the Anderson Darlin test

 

the ranking of DataFit is not an "must do", compare the graphs with your data and select the one, that is the closest to your data

Steffen Bangsow
freelance simulation specialist  
web: www.bangsow.eu
mail: steffen@bangsow.net

Re: Data fit tool issue

Siemens Phenom Siemens Phenom
Siemens Phenom

Hi Alessandro, hi Antonio,


if no theoretical distribution fits then you can use an empirical distribution. An example is explained in the Collection of small examples, which can be opened via the Start Page of Plant Simulation. Look at "DataFit" in the Example collection. The frame LookAt contains a method EmpiricalDistr.


You want to select single distributions then you go to the Fit tab button Select. You must press the Ctrl- key to select multiple distributions. The start the fitting process by the button Fit.
If the table Ranking (button Results on tab Evaluation) does not contains the selected distribution you can open the table messages on tab Evaluation. If there were errors during the calculation then you can find corresponding recordings.
If you want to fit data with values > 1 then the beta-distribution is impossible. You must filter the data on tab Data Filter, how it was mentioned by Steffen Bangsow.


Please note that in table Ranking there are results of a statistical test with a certain level of significance, which is defined on tab Fit. The columns like “Chi value” depends on this level of significance. The columns like “Chi statistic” depends on the sample. The test is successful if the value of “Chi statistic” < value of “Chi value”. You can decide if a small difference is acceptable or not.


How do you use a recommended Gumbel-distribution?
You must use it as a formula. The problem is that this distribution has also negative values.
I recommend using a method like this

 

-> real
var n:integer
repeat
result:= z_gumbel(1, 0.9, 1.1)
n += 1
if n > 20
throwRuntimeError("Impossible to generate the random number.")
end
until result>=0

Regards,

Peter

Re: Data fit tool issue

Creator
Creator
Thank you so much, Mr. Steffen, I have solved the problem. Thank for your reply.

Re: Data fit tool issue

Creator
Creator

Hi @Peter 

 

Thank you so much for your reply and your help to understand how to solve this.

 

Regards

Antonio