# Difference between revisions of "Tips"

From SUMOwiki

Line 7: | Line 7: | ||

* Remember that the Measure (and error function you use) strongly influence the quality and fit of the surrogate model. If you are unhappy with the final model, try a different [[Measures| Measure]] and/or [[FAQ#How_do_I_change_the_error_function_.28relative_error.2C_RMS.2C_....29.3F| error function]]. |
* Remember that the Measure (and error function you use) strongly influence the quality and fit of the surrogate model. If you are unhappy with the final model, try a different [[Measures| Measure]] and/or [[FAQ#How_do_I_change_the_error_function_.28relative_error.2C_RMS.2C_....29.3F| error function]]. |
||

− | * The [http://www.mathworks.com/products/neuralnet/ Matlab Neural Network Toolbox] is very slow! Try to avoid using |
+ | * The [http://www.mathworks.com/products/neuralnet/ Matlab Neural Network Toolbox] is very slow! Try to avoid using [[Measures#CrossValidation| CrossValidation]], but use a [[Measures#ValidationSet| ValidationSet]] instead (see the [[FAQ#Why_are_the_Neural_Networks_so_slow.3F | FAQ entry]]). |

− | |||

− | * If you use the RBF neural network model type and you get a crash in "[http://www.mathworks.com/access/helpdesk/help/toolbox/nnet/index.html?/access/helpdesk/help/toolbox/nnet/newrb.html&http://www.google.com/search?q=newrb&rls=com.microsoft:en-US:IE-SearchBox&ie=UTF-8&oe=UTF-8&sourceid=ie7&rlz=1I7GGIC newrb]" this is an error in the [http://www.mathworks.com/products/neuralnet/ Matlab Neural Network Toolbox] implementation and not anything we can do about (a workaround is available on [[Contact|request]]). This should be fixed by Matlab 7.5. |

## Revision as of 12:26, 6 June 2008

- If you want to benchmark your computer for Matlab speed simply run "bench" in Matlab

- By default Matlab only allocates about 117 MB memory space for the Java Virtual Machine. If you would like to increase this limit (which you should) please follow the instructions here. See also the general memory instructions here.

- You can switch off adaptive sample selection if you do not specify a <SampleSelector> tag. See Adaptive Modeling Mode.

- Remember that the Measure (and error function you use) strongly influence the quality and fit of the surrogate model. If you are unhappy with the final model, try a different Measure and/or error function.

- The Matlab Neural Network Toolbox is very slow! Try to avoid using CrossValidation, but use a ValidationSet instead (see the FAQ entry).