def dataGeneratorMain(args):
taskName = args.function
inputSize = None
outputSize = None
if args.function == 'ANN-bp':
inputSize = args.struct[0]
outputSize = args.struct[-1]
else:
inputSize = args.input_size
outputSize = args.output_size
dataSetSize = args.data_size
assert inputSize > 0 and inputSize <= 10
assert dataSetSize <= 23 and dataSetSize >= 3
#################################
# format the data set dir #
#################################
try:
# shell script to be run as subprocess
scriptFormatDir = """
#!/bin/bash
set -eu
# setup data set dir
orig_dir="`pwd`/"
dir_name={}
task_name="$0"
ip_size="$1"
set_size_pow="$2"
set_size_pow_start=0"$3"
if [ ! -d $orig_dir$dir_name ]
then
mkdir $orig_dir$dir_name
echo "created dir: $orig_dir$dir_name"
fi
if [ ! -d $orig_dir$dir_name$task_name ]
then
mkdir $orig_dir$dir_name$task_name
echo "created dir: $orig_dir$dir_name$task_name"
fi
echo "input size: "$ip_size
echo "set pow size: "$set_size_pow
task_dir=$orig_dir$dir_name$task_name
##################
cd $task_dir #
##################
existing_set="`ls`"
for size_pow in $(eval echo "{{$set_size_pow_start..$set_size_pow}}")
do
file_name=$size_pow
if [ "`find . -maxdepth 1 -type f -printf '%f\n' | grep $file_name`" ]
then
echo "data set already exists: "$file_name
else
touch $file_name
echo "created data set file: "$file_name
fi
done
##################
cd $orig_dir #
##################
""".format(trainingDirName)
# args to the script:
# $0: name for the training set (e.g.: sin: the data set should show characteristic of sin function)
# $1: number of input to the sigmoid neuron
# $2: indicate number of tuples in the data-set:
# range: 3 ~ 14
# e.g.:
# suppose user provide 10, then data set with size 2^3, 2^4, 2^5, ... 2^10 will be generated,
# and be stored as separate files in the corresponding folder
# $3: indicate min number of tuples in the data-set
if args.function == 'ANN-bp':
pass
else:
taskName += '_in-{}-out-{}'.format(inputSize, outputSize)
stdout, stderr = runScript(scriptFormatDir, [taskName, str(inputSize), str(dataSetSize), str(dataSetSizeStart)])
print("============================")
print("script msg: \n{}".format(stdout.decode('ascii')))
print("============================")
except ScriptException as se:
print(se)
#########################################
# generate data and write to file #
#########################################
genY = trainingFunc(args.function)
for dFile in os.listdir(trainingDirName + taskName):
dFileFull = trainingDirName + taskName + '/' + dFile
# all old files should already be read-only
if not (os.stat(dFileFull).st_mode & int('010010010',2)):
# i don't use os.access(file, os.W_OK) here, as it will always be true if you launch python3 as root in ec2
continue
if 'conf' in dFile or 'ignore' in dFile:
continue
dSize = dFile
dSize = int(dSize)
assert dSize >= dataSetSizeStart and dSize <= dataSetSize
assert os.stat(dFileFull).st_size == 0
f = open(dFileFull, 'w')
numEntry = pow(2, dSize)
for i in range(0, numEntry):
# randomly generate input list, within range 0 ~ 10
xList = [uniform(0, 10) for k in range(0, inputSize)]
yList = None
if args.function == 'ANN-bp':
pass
else:
yList = [genY(xList)]
dataList = yList + xList
dataStr = reduce(lambda x,y: str(x)+' '+str(y), dataList)
print(dataStr, file=f)
f.close()
##########################################################
# always enforce read-only policy for data set dir #
##########################################################
try:
scriptChmod = """
#!/bin/bash
orig_dir="`pwd`/"
dir_name={}
task_name="$0"
chmod 444 $orig_dir$dir_name$task_name/*
""".format(trainingDirName)
stdout, stderr = runScript(scriptChmod, [taskName])
except ScriptException as se:
print(se)
fi
done
##################
cd $orig_dir #
##################
"""
# args to the script:
# $0: name for the training set (e.g.: sin: the data set should show characteristic of sin function)
# $1: number of input to the sigmoid neuron
# $2: indicate number of tuples in the data-set:
# range: 3 ~ 14
# e.g.:
# suppose user provide 10, then data set with size 2^3, 2^4, 2^5, ... 2^10 will be generated,
# and be stored as separate files in the corresponding folder
# $3: indicate min number of tuples in the data-set
stdout, stderr = runScript(scriptFormatDir, [taskName, str(inputSize), str(dataSetSize), str(dataSetSizeStart)])
print("============================")
print("script msg: \n" + str(stdout))
print("============================")
except ScriptException as se:
print(se)
#########################################
# generate data and write to file #
#########################################
genY = trainingFunc(taskName)
for dFile in os.listdir(trainingDirName + taskName):
dFileFull = trainingDirName + taskName + '/' + dFile
# all old files should already be read-only
if not os.access(dFileFull, os.W_OK):