|
As Covean said, you can't use a 32bit dll form a 64bit application.
By default, a .NET application run according to the O.S. (so, in a 64bit systems run as 64bit application).
You have to force your application to run as 32bit by changing the "Platfor Target" value in the Build properties of your Visual Studio Project.
|
|
|
|
|
If you have a 64 bit .NET application attempting to run a 32 bit DLL, you can change it to run in 32 bit mode by using the corflags command. It's detailed here[^], but basically you need to run
corflags myassembly.exe /32bit+
"WPF has many lovers. It's a veritable porn star!" - Josh Smith As Braveheart once said, "You can take our freedom but you'll never take our Hobnobs!" - Martin Hughes.
My blog | My articles | MoXAML PowerToys | Onyx
|
|
|
|
|
that seems to be the only solution to avoid recompilation, many thanks!
Чесноков
|
|
|
|
|
You are welcome.
"WPF has many lovers. It's a veritable porn star!" - Josh Smith As Braveheart once said, "You can take our freedom but you'll never take our Hobnobs!" - Martin Hughes.
My blog | My articles | MoXAML PowerToys | Onyx
|
|
|
|
|
Hi,
When I Insert data in my Access Database show error i use this query
INSERT INTO trans (date,heads,debit,credit,descp) Values('2/8/2010','Jawad','0','9000','UNWASHD')
Table Trans
Feild....................Datatype
id......................AutoNumber
Date....................Datetime
Heads...................text
debit...................number
credit..................number
descp...................text
ERROR
"SYNTAX ERROR IN INSERT INTO STATEMENT "
Plz Help Me
Jawad Khatri
|
|
|
|
|
Date is a DateTime, you're treating it as a string.
debit & credit are numbers, you're treating bpth as a string.
Access, i think, uses # to denote a date in place of the quotes for strings, and numbers just dont have any quotes at all. hence:
INSERT INTO trans (date,heads,debit,credit,descp) Values(#2010-08-02#,'Jawad',0,9000,'UNWASHD')
|
|
|
|
|
Same Error 
|
|
|
|
|
I'm not sure about access but you shouldn't use columns with names like 'date' because this is can also be
a datatype in this database.
Greetings
Covean
|
|
|
|
|
And if you have to use such names, put them in square brackets [].
Next, use a parameterized query, that will also handle the date/time problems.
INSERT INTO [trans] ([date],[heads],[debit],[credit],[descp]) Values(?,?,?,?,?)'
|
|
|
|
|
Hi guys
I want to redicet a web page when user click on button in window based application using C#,
Like we do in asp.net "Response.Redirect("http://www.codeproject.com").
Is it possible and how
Thanks & Regards
Shafiq.....
|
|
|
|
|
Do you want something like
System.Diagnostics.Process.Start("www.codeproject.com");
|
|
|
|
|
Hey Goutam
Thank you. Its working fine.
Now i am asking little more if the page is already open and button click can we reload the same page
instead of opening new window. it will be helpful for me.
Thanks again
Shafiq
|
|
|
|
|
Hi, Everbody.
i have writen c# Artifical intelliget Neural Network. I have 500 count dataset which is normalized -1 and 1 and.i have used learning rate and momentum coefficient .
i teaching my network with 400 count dataset. Network have 5 inputs and 5 outputs. my output is setuped as 00001/00010/00100/01000/10000
when my network teched with 10000 iterations and 400 datasets, my output have a problem.
my network teached in last class of 10000 so when i tested my datasets , output only have been predicted as true class of 10000.
as result my code can only learn last class and other classes can not predicted or poor predicected
Example Test and Learning dataset
5 neuron input
{{0.04418 ,0.18135 ,0.28143 ,0.34367 ,0.14938 ,},
{0.37418 ,0.35050 ,0.18184 ,0.07493 ,0.01855 ,},
{0.04540 ,0.17061 ,0.24506 ,0.32560 ,0.21333 ,},
{0.08494 ,0.25653 ,0.26922 ,0.26141 ,0.12790 ,},
{0.18135 ,0.36832 ,0.24799 ,0.15743 ,0.04491 ,},
{0.11887 ,0.29143 ,0.27581 ,0.22870 ,0.08518 ,},
{0.02758 ,0.26239 ,0.37247 ,0.29998 ,0.03759 ,},
{0.00781 ,0.15792 ,0.32145 ,0.41811 ,0.09470 ,},
{0.07811 ,0.31608 ,0.31901 ,0.24506 ,0.04174 ,},
{0.07664 ,0.32390 ,0.32829 ,0.21674 ,0.05443 ,},
....
5 neuron output
{
{1 ,0 ,0 ,0 ,0 },
{1 ,0 ,0 ,0 ,0 },
{0 ,1 ,0 ,0 ,0 },
{0 ,1 ,0 ,0 ,0 },
{0 ,1 ,0 ,0 ,0 },
{0 ,0 ,1 ,0 ,0 },
{0 ,0 ,1 ,0 ,0 },
{0 ,0 ,0 ,1 ,0 },
{0 ,0 ,0 ,1 ,0 },
{0 ,0 ,0 ,0 ,1 },
{0 ,0 ,0 ,0 ,1 }
.....
My trainin code blok is as folowing show
int iterasyon=100000,it=0;
while (it <iterasyon && mmse >= 0.00001)
{
mmse = 0;
mae = 0;
for (int nf = 0; nf < 400; nf++)
{
mmse += hesaplas(ndata, nf);
trains(nf);
}
mmse = 0.5 * (mmse / 400 * 5));
if (it % 100 == 0)
{
yazs.WriteLine(mmse.ToString());
yazs.Flush();
}
it++;
}
modified on Tuesday, August 3, 2010 9:30 PM
|
|
|
|
|
Firstly, when you post code, use the "code block" widget rather than the "inline code" - it preserves the formatting and makes it easier to read:
int iterasyon=100000;
while (it = 0.00001)
{
mmse = 0;
mae = 0;
for (int nf = 0; nf < 400; nf++)
{
mmse += hesaplas(ndata, nf);
trains(nf);
}
mmse = 0.5 * (mmse / 400 * 5));
if (it % 100 == 0)
{
yazs.WriteLine(mmse.ToString());
yazs.Flush();
}
it++;
}
Secondly, if you are going to post code, make sure it compiles. That won't: it will complain about
while (it = 0.00001) "Cannot implicitly convert type xxx to bool"
Even if it did, it would execute the outer loop either once, or an infinite number of times.
Thirdly, don't use magic numbers: how do you know you have 400 entries? What is the
mmse = 0.5 * (mmse / 400 * 5));
supposed to do?
Real men don't use instructions. They are only the manufacturers opinion on how to put the thing together.
|
|
|
|
|
Three things (over Original Griffs Reply)
1. Is there any correlation beween the input an the output (I cannot see any)? if not, the NN is not likely to find it.
2. How are you training the network, your code is hard to read. If you are repeating outputs 1,0,0,0,0 (400/10000?) times at the end of trainging, you are probably annealing the network to give this result no matter what. You should cycle through many outputs.
3. It is possible to overtrain a NN!
I'd start with something simpler to test you NN code:
1 Input 0 -->360 , and train output to sin
1 Input 0 -->360 , and train output to x^2
2 Inputs, train to Logical Opertors (remember the XOR problem)
3 Inputs, result = i1 * i2 * i3
Then test on a real dataset.
ragnaroknrol The Internet is For Porn[^]
Pete o'Hanlon: If it wasn't insulting tools, I'd say you were dumber than a bag of spanners.
|
|
|
|
|
Firstly Thanks interest.
my dataset is folowing is that:
data is 5x400 matrix
b_cikis is 5x400 matrix
[...
long data removed
...].
int iterasyon=100000,it=0;
while (it = 0.00001)
{
mmse = 0;
mae = 0;
for (int nf = 0; nf < 400; nf++)
{
mmse += hesaplas(data, nf);
trains(nf);
}
mmse = 0.5 * (mmse / 400 * 5));
if (it % 100 == 0)
{
yazs.WriteLine(mmse.ToString());
yazs.Flush();
}
it++;
}
modified on Wednesday, August 4, 2010 10:25 AM
|
|
|
|
|
One of the admins has trimmed your message, the actual data is irrelevant and it was taking a long time to open your message.
I only want clarifiaction on two things, not the actual test data:
1. What does the input vector represent? What is its relationship to the output vector? The input looks like a bunch of random numbers to me.
2. Are you iterating 100000times over each input/output pair OR
Are you iterating over each pair of items in the input/ output arrays then doing this 100000 times.
ragnaroknrol The Internet is For Porn[^]
Pete o'Hanlon: If it wasn't insulting tools, I'd say you were dumber than a bag of spanners.
modified on Thursday, August 5, 2010 3:10 AM
|
|
|
|
|
Hi,
that is my actual daya set and i took my friend.
Datas is measured patient at hospital. As a result Datas is Biological datas, My Friend clasified nntool in matlab. he used difference learnin method which is Levenberg-Marquart, i used gradient desent with momentum algortihm(gdm). so i hope i will classify with my soft.
But test results is not good.
2.Yes
I train each pair in input/output arrays in each one iteration
for i:0:i<100000;i++
for j:0:j<400;j++
{
calculate error;
if error>0.000001
{ train network;
update weigths;
}
else
break;
}
|
|
|
|
|
OK the first thing we can rule out tis training to one input repeatedly at the end, which is a pity as it's the easiest fix  .
I think I understand the problem domain now, You are trying to categorise a function given 5 input points. The first thing is, you do have more than one layer between the input layer and the output layer don't you? If not you might hit the linear separability problem, as some of your functions look paraboloid, I *think* these aren't linearly separable, and would require at least two tiers of processing neurons
In your OP You defined the trainging outputs to be:
2 X Category I input
3 X Category II input (is this correct? the others all take two)
2 x Category II input.
I stopped here. I then entered your test data into excel (red = Cat I, green = Cat II Blue = Cat III) and plotted their "function" graphs
{0.04418 ,0.18135 ,0.28143 ,0.34367 ,0.14938}
{0.37418 ,0.35050 ,0.18184 ,0.07493 ,0.01855}
{0.04540 ,0.17061 ,0.24506 ,0.32560 ,0.21333}
{0.08494 ,0.25653 ,0.26922 ,0.26141 ,0.12790}
{0.18135 ,0.36832 ,0.24799 ,0.15743 ,0.04491}
{0.11887 ,0.29143 ,0.27581 ,0.22870 ,0.08518}
{0.02758 ,0.26239 ,0.37247 ,0.29998 ,0.03759}
I suggest you do the same. AFAICT plots for 1 and 3 are near identical (despite being in different categories, as do plots 4 & 6), plots 1 & 2 are different (depsite being in the same category, plot 1 looks like the maximun of a quadratic, the other is sigmoid.), this is also the case for graphs 6 & 7. If your input data is correct, it is doubtful than at NN can train to these sets.
The situation looks a little better if the expected outputs are changed to
2 X Category I input
2 X Category II input (corrected?)
2 x Category II input.
In theory what you are doing is similar to OCR, but the graphs do not look consistent enough within a category, or different enough between categories for you approach to work. Hopefully I'm wrong though!!!!!!
ragnaroknrol The Internet is For Porn[^]
Pete o'Hanlon: If it wasn't insulting tools, I'd say you were dumber than a bag of spanners.
|
|
|
|
|
OP (Karayel) e-mailed me directly with the following
> Hi ,
>
> i have 5 layer and have 3 hiden layer in network
> datas is not mine, i said that i took from my friends Lab. datas is measure from patients.
> they can be wrong.
>
> i have 5 category in my data set
> i am sending you my datas
[edit removed the data set!]
The fact that you have 3 hidden layers will solve the linear separability problem.
Actually, this is the crux of your problem:
"datas is measure[d] from patients. they can be wrong"
Firstly, I replotted the first three graphs of category one. The graphs one an three looked very similar (a line going up hitting a maximum, then going down again)(which is good), but graph 2 looked sigmoid (it could have been the same graph but shifted earlier).The problem is this breaks the pattern, and I suspect this happens in all the categories, if so the NN is unlikely to be any help IMO. People have tried to spot predict patterns in stockmarket prices with limited success, I think the situation you have is similar.
If the graphs are consistent (and you should plot samples from each category yourself to check) you may have a hope. If this is the case there are three things I'd Add
1) if this is a research project, is a NN really suitable in vivo? The system you are writing is inherently unprovable, and given this is a medical diagnostic tool (with potentially serious consequences), should it be trusted?
2) You are training lots of category 1 inputs, then lots of category 2,etc. If your NN anneals, it could well anneal to category 1 by doing it this way. You should cycle through one example of Cat I, Cat II Cat III etc while training, and repeat until the NN is as accurate as you need it. If you can develop a function to simulate expected data sets, that would be even better, as real data is unreliable, and trains unreliably. Ironically, it is the unreliable input that the NN sorts out, but I'd use generate training inputs.
3) Can you express what constitues category 1,2,3, etc formally in code? If so just coding it will be better.
ragnaroknrol The Internet is For Porn[^]
Pete o'Hanlon: If it wasn't insulting tools, I'd say you were dumber than a bag of spanners.
|
|
|
|
|
Hi i am writing again.
I want to test my code and
is there any fucntion which are 5 inputs and 5 out?
or square function example, 2 inputs x 2 outputs, 4 inputs x4 outputs.
May be , you can recommend a function for my testing.
in addition, i thank you so mouch for sharining your informations with me
Best regards.
|
|
|
|
|
If your are just testing the NN works, I'd lower the inputs and outputs to 1 neuron, then train to some functions (sin, x^2 ln etc), this gives you a base line to ensure all is correct.
As for 5 inputs, 5 outputs, I'm not sure, the problem I applied my NN to was quite different. You could try a matrix transform of some sort, or categories 5 inputs. You might have better luck contacting a dedicated NN board (I'm sure I've seen them somewhere).
I have only used a NN once(for my MSc Dissertation) and that was ~10 years ago so I'm pretty rusty. Every couple of years I get the urge go back to it, but other things are shinier (and /or more remunerative . I had to hand-check my NN there was a term in the momentum calculation (actually a "-" instead of a "+"!) that brought the NN close to the desired result, then would diverge again. That took two weeks (c++ on linux, nedit, but no real IDE).
Anyway, best of luck.
ragnaroknrol The Internet is For Porn[^]
Pete o'Hanlon: If it wasn't insulting tools, I'd say you were dumber than a bag of spanners.
|
|
|
|
|
tanks you too,
i used meumentum coefficient.
May be you will begin study NN again?
can you see your code ? and in there, can you write main block in your code?
best.
|
|
|
|
|
karayel_kara wrote: I used meumentum coefficient.
It's a good thing to use, I tested the property my NN without the coefficient, and learning was much slower, I just had a bug in mine!
karayel_kara wrote: May be you will begin study NN again?
Not likely sadly, not much call for it in the business world and most of what I do is related to business.
karayel_kara wrote: can you see your code ? and in there, can you write main block in your code?
If memory serves ~2000+ lines of [badly written, I had only just started developing!] c++, and I only have a hardcopy here, the CD is back in the UK. There are neural network frameworks available on CP. This one is really good:
http://www.codeproject.com/KB/recipes/aforge_neuro.aspx[^]
ragnaroknrol The Internet is For Porn[^]
Pete o'Hanlon: If it wasn't insulting tools, I'd say you were dumber than a bag of spanners.
|
|
|
|
|
hye all,
How do I mask my image such that i got the masked image. For example i want to mask round shape and discard the background. Anyone could guide me?
|
|
|
|
|
General 
News 
Suggestion 
Question 
Bug 
Answer 
Joke 
Praise 
Rant 
Admin 
Submit an article or tip
