def test_calcMean(self):
modelMetadata = TestModel_Metadata()
model = TestModel(modelMetadata)
smoothingMetadata = dc.DefaultSmoothingOscilationEWMA_Metadata(
movingAvgParam=0.5)
smoothing = dc.DefaultSmoothingOscilationEWMA(
smoothingMetadata=smoothingMetadata)
smoothing.__setDictionary__(
smoothingMetadata=smoothingMetadata,
dictionary=model.getNNModelModule().named_parameters())
smoothing.calcMean(model=model, smoothingMetadata=smoothingMetadata)
wg = dict(model.named_parameters())
smoothedWg = smoothing.weightsSum
self.compareDictToNumpy(wg, init_weights)
self.compareDictToNumpy(smoothedWg, init_weights)
#############
smoothing.calcMean(model=model, smoothingMetadata=smoothingMetadata)
wg = dict(model.named_parameters())
smoothedWg = smoothing.weightsSum
self.compareDictToNumpy(wg, init_weights)
self.compareDictToNumpy(smoothedWg, init_weights)
############
second_base_weights = {
'linear1.weight': [[17., 17., 17.]],
'linear1.bias': [19.],
'linear2.weight': [[17.], [17.], [17.]],
'linear2.bias': [19., 19., 19.]
}
second_smth_weights = {
'linear1.weight': [[11., 11., 11.]],
'linear1.bias': [13.],
'linear2.weight': [[11.], [11.], [11.]],
'linear2.bias': [13., 13., 13.]
}
model.setConstWeights(weight=17, bias=19)
smoothing.calcMean(model=model, smoothingMetadata=smoothingMetadata)
wg = dict(model.named_parameters())
smoothedWg = smoothing.weightsSum
self.compareDictToNumpy(iterator=wg, numpyDict=second_base_weights)
self.compareDictToNumpy(iterator=smoothedWg,
numpyDict=second_smth_weights)
dataMetadata = dc.DefaultData_Metadata(pin_memoryTest=False,
pin_memoryTrain=False,
epoch=2,
fromGrayToRGB=True)
loop = 5
modelName = "wide_resnet50_2"
prefix = "epsilons_"
types = ('predefModel', 'MNIST', 'movingMean')
try:
stats = []
rootFolder = prefix + sf.Output.getTimeStr() + ''.join(
x + "_" for x in types) + "set"
smoothingMetadata = dc.DefaultSmoothingOscilationEWMA_Metadata(
movingAvgParam=0.15,
epsilon=1e-1,
hardEpsilon=1e-3,
weightsEpsilon=1e-2,
batchPercentMaxStart=0.98)
for r in range(loop):
obj = models.wide_resnet50_2()
metadata.resetOutput()
modelMetadata = dc.DefaultModel_Metadata()
stat = dc.run(numbOfRepetition=2,
modelType=types[0],
dataType=types[1],
smoothingType=types[2],
metadataObj=metadata,
modelMetadata=modelMetadata,
dataMetadata=dataMetadata,
pin_memoryTrain=False,
epoch=2,
fromGrayToRGB=True)
loop = 5
modelName = "alexnet"
prefix = "mov_param_"
runningAvgSize = 10
types = ('predefModel', 'MNIST', 'movingMean')
try:
stats = []
rootFolder = prefix + sf.Output.getTimeStr() + ''.join(
x + "_" for x in types) + "set"
smoothingMetadata = dc.DefaultSmoothingOscilationEWMA_Metadata(
movingAvgParam=0.05,
epsilon=1e-5,
hardEpsilon=1e-7,
weightsEpsilon=1e-6,
batchPercentMaxStart=0.98)
for r in range(loop):
obj = models.alexnet()
metadata.resetOutput()
modelMetadata = dc.DefaultModel_Metadata()
data = DataMap[dataType](dataMetadata)
smoothing = SmoothingMap[smoothingType](smoothingMetadata)
model = ModelMap[modelType](obj=modelPredefObj,
modelMetadata=modelMetadata,
name=modelPredefObjName,
optimizer=optimizer,
lossFunc=lossFunc)