|
|
| rohit ghosh 2005-11-09, 8:00 am |
| hi everybody, i m doing a project in neural networks of stock
prediction in matlab. i m able to train the system but not able to
predict and test the training. please help me in this regard.
| |
| Greg Heath 2005-11-10, 7:06 pm |
|
rohit ghosh wrote:
> hi everybody, i m doing a project in neural networks of stock
> prediction in matlab. i m able to train the system
What does that mean?
>but not able to predict and test the training.
What does that mean?
> please help me in this regard.
Why do you ask for help without giving any useful information?
1. How many cases do you have?
2. How many cases must be reserved for testing?
3. How many input variables?
4. How many output variables?
5. How far do you want to predict into the future?
6. How far into the past do you want to base the prediction?
7. Have you designed and tested a linear model before
trying to tackle a neural net?
Hope this helps.
Greg
| |
| Greg Heath 2005-11-10, 7:06 pm |
|
Gabe wrote:
> rohit ghosh wrote:
>
> I'm not sure how you can train a NN (and know that it is converging
> properly) without knowing what the output of the NN is. That output
> should be the prediction you are looking for.
>
> In order to train and test your NN, you split your data set into a
> couple of pieces. One portion is used to train the NN and the other
> is used to test the NN. I don't know what the standard split is but
> if you have 10000 examples then you could train with 5000 and test
> with the other 5000. You have to be careful about how you choose
> your train and test data sets though. Both sets should, especially
> the training set, include the full range of possible inputs and
> outputs.
Not very applicable for predicting into the future.
Hope this helps.
Greg
| |
| Greg Heath 2005-11-10, 7:06 pm |
|
Gabe wrote:
> rohit ghosh wrote:
>
> I'm not sure how you can train a NN (and know that it is converging
> properly) without knowing what the output of the NN is. That output
> should be the prediction you are looking for.
>
> In order to train and test your NN, you split your data set into a
> couple of pieces. One portion is used to train the NN and the other
> is used to test the NN. I don't know what the standard split is but
> if you have 10000 examples then you could train with 5000 and test
> with the other 5000. You have to be careful about how you choose
> your train and test data sets though. Both sets should, especially
> the training set, include the full range of possible inputs and
> outputs.
Not very applicable for predicting into the future.
Hope this helps.
Greg
| |
| Steve Sipes 2005-12-18, 9:59 pm |
| I am also interested in stock market (Quant/black box ) investing ..
I have collected historical data points that I would like to test.
However.. I dont know where to start on this .. I have an excel file
that my data is on .. can I import the excel file into matlab?? I am
using volume data now ..was wondering which would be better ..(
either volume data or daily stock price) in my NN. I guess I also
need to know which network would be best to run test this on.. any
pointers? any suggestions?? ..So first things first .. can I import
my excel file into matlab... ?
Thanks
Steve -
Greg Heath wrote:
>
>
>
> rohit ghosh wrote:
>
> What does that mean?
>
>
> What does that mean?
>
>
> Why do you ask for help without giving any useful information?
>
> 1. How many cases do you have?
> 2. How many cases must be reserved for testing?
> 3. How many input variables?
> 4. How many output variables?
> 5. How far do you want to predict into the future?
> 6. How far into the past do you want to base the prediction?
> 7. Have you designed and tested a linear model before
> trying to tackle a neural net?
>
> Hope this helps.
>
> Greg
>
>
|
|
|
|