Prev: Expectation of a matrix..please help me it s urgent
Next: cubes inside a sphere
From: Swapnil on 6 Apr 2010 03:45
I have designed neural network for intrusion detection system.
I am using KDD 99 dataset to train the network. It has around 4,00,000 records.
When I apply records to the designed network, MATLAB hangs at any time......
It has no pattern about this behaviour.... sometimes it hangs at the start of training, sometimes at the last (no any observable pattern)........
I am using FEDORA 10. & MATLAB 7.R.14
Please help m out.....
From: Greg Heath on 6 Apr 2010 09:32
On Apr 6, 3:45 am, "Swapnil " <gujarathiswap...(a)gmail.com> wrote:
> I have designed neural network for intrusion detection system.
> I am using KDD 99 dataset to train the network. It has around 4,00,000 records.
> When I apply records to the designed network, MATLAB hangs at any time.......
> It has no pattern about this behaviour.... sometimes it hangs at the start of training, sometimes at the last (no any observable pattern)........
> I am using FEDORA 10. & MATLAB 7.R.14
> Please help m out.....

Thanks for all of the information.

Error messages?

Do you really need Ntrn ~ 4e+6??
Network size I:H:O = ?
Training Algorithm?
Nondefault training parameters?

Greg
From: Swapnil on 7 Apr 2010 03:23
Greg Heath <heath(a)alumni.brown.edu> wrote in message <dc392b17-e84b-47d3-bacd-756f63ece92e(a)z7g2000yqb.googlegroups.com>...
> On Apr 6, 3:45 am, "Swapnil " <gujarathiswap...(a)gmail.com> wrote:
> > I have designed neural network for intrusion detection system.
> > I am using KDD 99 dataset to train the network. It has around 4,00,000 records.
> > When I apply records to the designed network, MATLAB hangs at any time......
> > It has no pattern about this behaviour.... sometimes it hangs at the start of training, sometimes at the last (no any observable pattern)........
> > I am using FEDORA 10. & MATLAB 7.R.14
> > Please help m out.....
>
> Thanks for all of the information.
>
> Error messages?
>
> Do you really need Ntrn ~ 4e+6??
> Network size I:H:O = ?
> Training Algorithm?
> Nondefault training parameters?
>
> Greg

> Error messages?
I dont get any error message on the promt but it generates blank matlab_crash_dump.XXXX file (XXXX- PID). It shows the blank screen and unable to process any further.... I have to kill that matlab process and resume.

> Network size I:H:O = ?
> Training Algorithm?
I am using winner take all algorithm. There are 3 layers of neurons and each layer has 36 neurons. All the layers are trained using the same algorithm......
I also tried with lesser number of neurons (23) but still is the issue......
output of each neuron is a cluster which contains records of similar attack type.

Out of 4,00,000 records i have created subset of 43000 records and applying for training.......

> Nondefault training parameters?
I didnt get its meaning..... I would really appreciate if u could elaborate it....

thnx.
From: Greg Heath on 7 Apr 2010 22:07
On Apr 7, 3:23 am, "Swapnil " <gujarathiswap...(a)gmail.com> wrote:
> Greg Heath <he...(a)alumni.brown.edu> wrote in message <dc392b17-e84b-47d3-bacd-756f63ece...(a)z7g2000yqb.googlegroups.com>...
> > On Apr 6, 3:45 am, "Swapnil " <gujarathiswap...(a)gmail.com> wrote:
> > > I have designed neural network for intrusion detection system.
> > > I am using KDD 99 dataset to train the network. It has around 4,00,000 records.
> > > When I apply records to the designed network, MATLAB hangs at any time......
> > > It has no pattern about this behaviour.... sometimes it hangs at the start of training, sometimes at the last (no any observable pattern).........
> > > I am using FEDORA 10. & MATLAB 7.R.14
> > > Please help m out.....
>
> > Thanks for all of the information.
>
> > Error messages?
>
> > Do you really need Ntrn ~ 4e+6??
> > Network size I:H:O = ?
> > Training Algorithm?
> > Nondefault training parameters?
>
> > Greg
> > Error messages?
>
> I dont get any error message on the promt but it generates blank matlab_crash_dump.XXXX  file (XXXX- PID). It shows the blank screen and unable to process any further.... I have to kill that matlab process and resume.
>
> > Network size I:H:O = ?
> > Training Algorithm?
>
> I am using winner take all algorithm. There are 3 layers of neurons and each layer has 36 neurons. All the layers are trained using the same algorithm......
> I also tried with lesser number of neurons (23) but still is the issue.......
> output of each neuron is a cluster which contains records of similar attack type.
>
> Out of 4,00,000 records i have created subset of 43000 records and applying for training.......  
>
> > Nondefault training parameters?
>
> I didnt get its meaning..... I would really appreciate if u could elaborate it....
>
> thnx

Most of the NN Toolbox algorithms have default parameters
which can be overwritten.

1. What algorithm are you using?
2. What default parameters are you overwriting?

Greg
From: Swapnil on 8 Apr 2010 02:47
Greg Heath <heath(a)alumni.brown.edu> wrote in message <9f87b518-fa94-43d4-83c1-ccee0c0420aa(a)z4g2000yqa.googlegroups.com>...
> On Apr 7, 3:23 am, "Swapnil " <gujarathiswap...(a)gmail.com> wrote:
> > Greg Heath <he...(a)alumni.brown.edu> wrote in message <dc392b17-e84b-47d3-bacd-756f63ece...(a)z7g2000yqb.googlegroups.com>...
> > > On Apr 6, 3:45 am, "Swapnil " <gujarathiswap...(a)gmail.com> wrote:
> > > > I have designed neural network for intrusion detection system.
> > > > I am using KDD 99 dataset to train the network. It has around 4,00,000 records.
> > > > When I apply records to the designed network, MATLAB hangs at any time......
> > > > It has no pattern about this behaviour.... sometimes it hangs at the start of training, sometimes at the last (no any observable pattern)........
> > > > I am using FEDORA 10. & MATLAB 7.R.14
> > > > Please help m out.....
> >
> > > Thanks for all of the information.
> >
> > > Error messages?
> >
> > > Do you really need Ntrn ~ 4e+6??
> > > Network size I:H:O = ?
> > > Training Algorithm?
> > > Nondefault training parameters?
> >
> > > Greg
> > > Error messages?
> >
> > I dont get any error message on the promt but it generates blank matlab_crash_dump.XXXX  file (XXXX- PID). It shows the blank screen and unable to process any further.... I have to kill that matlab process and resume.
> >
> > > Network size I:H:O = ?
> > > Training Algorithm?
> >
> > I am using winner take all algorithm. There are 3 layers of neurons and each layer has 36 neurons. All the layers are trained using the same algorithm......
> > I also tried with lesser number of neurons (23) but still is the issue......
> > output of each neuron is a cluster which contains records of similar attack type.
> >
> > Out of 4,00,000 records i have created subset of 43000 records and applying for training.......  
> >
> > > Nondefault training parameters?
> >
> > I didnt get its meaning..... I would really appreciate if u could elaborate it....
> >
> > thnx
>
> Most of the NN Toolbox algorithms have default parameters
> which can be overwritten.
>
> 1. What algorithm are you using?
> 2. What default parameters are you overwriting?
>
> Greg

I am not using any NN toolbox. I have implemented the winner take all algorithm which has a user defined parameters such as learning rate, number of neurons in each layer (as I said in earlier post) , number of layers.........

Swapnil.
 |  Next  |  Last
Pages: 1 2
Prev: Expectation of a matrix..please help me it s urgent
Next: cubes inside a sphere