run.bat

基于稀疏网络的精选机器学习模型

@echo "Testing SNoW using the boolean example data..."
@echo ""

@echo off

REM SNoW usage example: Training on simple boolean examples
REM
REM The training examples used for this example are the same as those given in
REM the user guide for the disjunction x2 v x5.  Using 9 examples of the
REM disjunction and two cycles over these examples, SNoW learns the concept with
REM 93.75% accuracy after two rounds of training.  In a simple boolean example
REM like this, it turns out that a single Winnow learner will perform better
REM than two Winnows in the SNoW architecture.  To confirm this, uncomment both
REM of the "set" commands that are commented out below and run this script
REM again.  Since only one target is specified in the one_target.arch
REM architecture file, SNoW will make predictions as classical Winnow does:
REM comparing the target's activation to the specified threshold.  With the
REM parameters specified in one_target.arch, classical Winnow learns the concept
REM with 100% accuracy over the entire six feature space after two rounds of
REM training.

REM Note the use of an eligibility threshold of 1.  This is important in small
REM experiments such as this one, where some important features are much more
REM likely to only appear once with a target in the training examples.  Also
REM note the -g- option, which disables the automatic generation of
REM conjunctions.  Finally, note the use of the "allactivations" output mode,
REM allowing us to see results at a more detailed level than the default.

REM Set this variable to match the path to your executable file
set snow=snow.exe

set train_file=two_targets_train.snow
REM set train_file=one_target_train.snow
set test_file=booleantest.snow

set net_file=boolean.net
set arch_file=two_targets.arch
REM set arch_file=one_target.arch
set results_file=boolean.out

@echo on

%snow% -train -I %train_file% -F %net_file% -A %arch_file% -e count:1 -g-
@echo ""
%snow% -test -I %test_file% -F %net_file% -o allboth -R %results_file%