Next: Further Information
Up: Network Classifier
Previous: Training a network
Contents
Adding New Types Of Parameters
An attempt has been made to make it as straightforward as possible for
the user to add new types of parameters to the network input. This is done
by carrying out the following steps.
- The TCLN 1 titles bank should be extended to hold an entry for each
new parameter type. The bank has been designed for these new entries to be
conveniently added to the end. For single parameters the control word
just specifies whether the parameter is to be used or not (1 or 0). For
parameter sets (eg harmonic) the word specifies how many of that
parameter type are to be calculated. If a parameter type produces a
fixed number of parameters which is greater than 1, then the user must
specify the exact number of parameters, and CLN_CLASSIFIER_INI
should check that only that number is enabled, and should reject
non-valid values.
- The KCLN_PAR_TYPES parameter in the include file CLN_COM.INC
must be changed to give the total number of parameter types (it is used
to step through each parameter type in TCLN 1 and count the total number of
parameters required.) It is also a good idea to create a new titles bank
mnemonic for each new parameter type in TCLN 1.
- The routine CLN_GET_PARAM must be extended to call the user's
own routines which calculate the new parameters. This involves the
addition of a few simple lines which should follow the format of the
examples in that routine. Note that the order of obtaining the
parameters is the order in which the CLNP parameters bank is filled
with those parameters. To avoid confusion it is advisable to keep the
ordering here in line with that in the TCLN 1 titles bank.
- Finally, the user should provide routines to actually calculate
the new parameters. These can make use of the pmt array (hits passing
residual cut) and the fitted vertex or direction, all of which are
accessed as call arguments. The use of the pmt array is optional for
those parameter types which do not require it, or which use all tubes
in an event. The user may also book and output diagnostic histograms
(see the ACF parameters code for an example of this).
Next: Further Information
Up: Network Classifier
Previous: Training a network
Contents
sno Guest Acct
2009-09-09