nn_var

快速傅立叶变换程序代码,学信号的同学,可要注意了

# neural net variables
#
# file to be read by commander.p
# each line is turned into a structure entry   	(str)
#     	a default-setting entry			(def)
#     	a usage-printing entry			(usg)
# 	and a command line reading entry	(clr)
# of these, the S, U and C can be disabled by >No S, etc.
# and the D can be disabled by -.
#
# variable    type deflt flag  short_name extra_action long_name
#
nc->report 	d  1 	-nnr report 		-	reporting style
nc->verbose 	d  1 	-nnv verbose		-	verbosity 0/1/2
nc->decverbose 	d  1 	-nndv decverbose	-	verbosity in decoding
nc->decwrite 	d  0 	-nndwrite verbosity	-	write decoding info to file
nc->decfile	s  - 	-nndecfile file nc->decwrite=2; file for decoding info
#
>PU   NLNE; fprintf( fp, " Neural net initialization:");
>PS  /* Neural net initialization */
#
#  int write ; /* if write == 2 then default outfile is overridden */
#  int read ;  /* if read  == 2 then default infile is overridden */
#
nc->read        d  0   	-nnread read	 	-	whether to read wts
nc->infile    	s  -    -nnin infile   	nc->read=2;  	weights from (instead of default)
nc->outfile    	s  -    -nnout outfile  nc->write=2;  	weights out (instead of default)
nc->init_rule  	d  1 	-nninit rule		-	how to init wts
nc->def_w 	f 1.0 -nndef_w  def_w		-	default initial weight
nc->def_b 	f 0.0 -nndef_b def_b		- 	default initial bias
nc->sigma_w0 	f  0.3 	-nnsigmaw0 sigma	-	initial random wts
nc->wseed	ld 2489 -nnwseed wseed		-	weight randomization
#
>PU   NLNE; fprintf( fp, " Neural net training:");
>PS  /* 	About training */
#
nc->train	d  0 	-nntrain train		-	whether to train
nc->train_n	d  100 	-nnn n 			-	training number
nc->test_n	d  1000	-nntn n			-	training number
nc->trseed	ld 4896 -nntrseed trseed 	-	defines training set
nc->teseed	ld 126999 -nnteseed teseed 	-	test set
nc->regularize  d  1	-nnregularize	r	-	type of regularization 0/1/2
>No S
nc->alpha[1] 	f  0.000001  -nna1 a1 		-	regularization of bias
nc->alpha[2] 	f 0.01	     -nna2 a2	 	-	regularization of inps
nc->alpha[3] 	f 20.0       -nna3 a3	 	-	regularization of 2nd type inps
>S
#   regularization constants:					
#   bias is 1, inputs are 2 					
#   initial runs gave sigma1 = 35 and s2 = 0.31	 	
#
>PU  NLNE; fprintf( fp, " Neural net optimizer:");
>PS  /* Optimization procedure */
#
nc->opt		d 2 -nnopt opt			- macopt1 or 2
nc->LOOP	d 5 -nnloops loops		-	Number of macopt runs
nc->itmax 	d 100 -nnitmax itmax 		-	max no line searches
nc->tolmin 	f 0.00001 -nntolmin tolmin	- final tolerance in training
nc->tol0 	f 0.1  -nntol0 tol0		-	initial tolerance in training
nc->rich 	d 0  -nnrich rich		-	expensive optimizer?
nc->end_on_step d 1  -nneos eos			- termination condition is that step is small
nc->CG		d 0  -cg cg			- whether to check gradient, on how many
nc->epsilon 	f 0.0001 -nneps epsilon 	- epsilon for check gradient
nc->evalH 	d 1  	-nnevalH evalH		- evaluate hard performance measures
#
>PU  NLNE; fprintf( fp, " Neural net decoding procedure:");
>PS  /* Neural net decoding procedure */
#
nc->hitlist_policy d 2 	-nnhp hp			-	1=if threshold exceeded; 2=sort 
nc->hitlist_thresh f 0.99 -nnhpt t  	   	-	hitlist threshold
nc->hitlist_n	d 10 	-nnhpn	n		-	number to aim to hit 
nc->hitlist_low  f 0.5  -nnhpl l		- 	-
nc->decodits	d 10 	-nndecodits its		-	max number of iterations to do when decoding
nc->decodn	d 1000 	-nndecodn n		-	number of examples to try to decode
nc->decodseed	ld 126999 -nndecodseed seed	-	seed for decoding tests
#
>PS  /* various leftovers */
#
>No CU
nc->write	d  1  -		 -		-	whether to write weights
nc->writeit 	d  0  	- -		 	-	undefined weight writing flag
nc->RC 		d 2   - -			-  	no of reg classes
nc->tolf 	f 0.5 - - 			-	factor by which tolerance decreases
nc->tol 	f 0.01 - -			-	tolerance in training
nc->itEH	d -1 - - -				iterative decoder's error
nc->itEHwb	d -1 - - -				iterative decoder's error (whole blocks)
>CU
>No S
net->thresh 	f 0.5 -nnthresh thresh		-	hard decision boundary
>S