def createNetwork(self, networkName, folderName, cnnLayers, kernel_Shapes,
intermediate_ConnectedLayers, n_classes,
sampleSize_Train, sampleSize_Test, batch_Size,
applyBatchNorm, numberEpochToApplyBatchNorm,
activationType, dropout_Rates, pooling_Params,
weights_Initialization_CNN, weights_Initialization_FCN,
weightsFolderName, weightsTrainedIdx, softmax_Temp):
# ============= Model Parameters Passed as arguments ================
# Assign parameters:
self.networkName = networkName
self.folderName = folderName
self.cnnLayers = cnnLayers
self.n_classes = n_classes
self.kernel_Shapes = kernel_Shapes
self.intermediate_ConnectedLayers = intermediate_ConnectedLayers
self.pooling_scales = pooling_Params
self.dropout_Rates = dropout_Rates
self.activationType = activationType
self.weight_Initialization_CNN = weights_Initialization_CNN
self.weight_Initialization_FCN = weights_Initialization_FCN
self.weightsFolderName = weightsFolderName
self.weightsTrainedIdx = weightsTrainedIdx
self.batch_Size = batch_Size
self.sampleSize_Train = sampleSize_Train
self.sampleSize_Test = sampleSize_Test
self.applyBatchNorm = applyBatchNorm
self.numberEpochToApplyBatchNorm = numberEpochToApplyBatchNorm
self.softmax_Temp = softmax_Temp
# Compute the CNN receptive field
stride = 1
self.receptiveField = computeReceptiveField(self.kernel_Shapes, stride)
# --- Size of Image samples ---
self.sampleSize_Train = sampleSize_Train
self.sampleSize_Test = sampleSize_Test
## --- Batch Size ---
self.batch_Size = batch_Size
# ======== Calculated Attributes =========
self.centralVoxelsTrain = getCentralVoxels(self.sampleSize_Train,
self.receptiveField)
self.centralVoxelsTest = getCentralVoxels(self.sampleSize_Test,
self.receptiveField)
#==============================
rng = numpy.random.RandomState(23455)
# Transfer to LIVIA NET
self.sampleSize_Train = sampleSize_Train
self.sampleSize_Test = sampleSize_Test
# --------- Now we build the model -------- #
print("...[STATUS]: Building the Network model...")
# Define the symbolic variables used as input of the CNN
# start-snippet-1
# Define tensor5
tensor5 = T.TensorType(dtype='float32',
broadcastable=(False, False, False, False,
False))
self.inputNetwork_Train = tensor5()
self.inputNetwork_Test = tensor5()
# Define input shapes to the netwrok
inputSampleShape_Train = (self.batch_Size, 1, self.sampleSize_Train[0],
self.sampleSize_Train[1],
self.sampleSize_Train[2])
inputSampleShape_Test = (self.batch_Size, 1, self.sampleSize_Test[0],
self.sampleSize_Test[1],
self.sampleSize_Test[2])
print(" - Shape of input subvolume (Training): {}".format(
inputSampleShape_Train))
print(" - Shape of input subvolume (Testing): {}".format(
inputSampleShape_Test))
inputSample_Train = self.inputNetwork_Train
inputSample_Test = self.inputNetwork_Test
# TODO change cnnLayers name by networkLayers
self.generateNetworkLayers(cnnLayers, kernel_Shapes,
self.pooling_scales, inputSampleShape_Train,
inputSampleShape_Test, inputSample_Train,
inputSample_Test,
intermediate_ConnectedLayers)
def createNetwork(self,
networkName,
folderName,
cnnLayers,
kernel_Shapes,
intermediate_ConnectedLayers,
n_classes,
sampleSize_Train,
sampleSize_Test,
batch_Size,
applyBatchNorm,
numberEpochToApplyBatchNorm,
activationType,
dropout_Rates,
pooling_Params,
weights_Initialization_CNN,
weights_Initialization_FCN,
weightsFolderName,
weightsTrainedIdx,
softmax_Temp
):
# ============= Model Parameters Passed as arguments ================
# Assign parameters:
self.networkName = networkName
self.folderName = folderName
self.cnnLayers = cnnLayers
self.n_classes = n_classes
self.kernel_Shapes = kernel_Shapes
self.intermediate_ConnectedLayers = intermediate_ConnectedLayers
self.pooling_scales = pooling_Params
self.dropout_Rates = dropout_Rates
self.activationType = activationType
self.weight_Initialization_CNN = weights_Initialization_CNN
self.weight_Initialization_FCN = weights_Initialization_FCN
self.weightsFolderName = weightsFolderName
self.weightsTrainedIdx = weightsTrainedIdx
self.batch_Size = batch_Size
self.sampleSize_Train = sampleSize_Train
self.sampleSize_Test = sampleSize_Test
self.applyBatchNorm = applyBatchNorm
self.numberEpochToApplyBatchNorm = numberEpochToApplyBatchNorm
self.softmax_Temp = softmax_Temp
# Compute the CNN receptive field
stride = 1;
self.receptiveField = computeReceptiveField(self.kernel_Shapes, stride)
# --- Size of Image samples ---
self.sampleSize_Train = sampleSize_Train
self.sampleSize_Test = sampleSize_Test
## --- Batch Size ---
self.batch_Size = batch_Size
# ======== Calculated Attributes =========
self.centralVoxelsTrain = getCentralVoxels(self.sampleSize_Train, self.receptiveField)
self.centralVoxelsTest = getCentralVoxels(self.sampleSize_Test, self.receptiveField)
#==============================
rng = numpy.random.RandomState(23455)
# Transfer to LIVIA NET
self.sampleSize_Train = sampleSize_Train
self.sampleSize_Test = sampleSize_Test
# --------- Now we build the model -------- #
print("...[STATUS]: Building the Network model...")
# Define the symbolic variables used as input of the CNN
# start-snippet-1
# Define tensor5
tensor5 = T.TensorType(dtype='float32', broadcastable=(False, False, False, False, False))
self.inputNetwork_Train = tensor5()
self.inputNetwork_Test = tensor5()
self.inputNetwork_Train_Bottom = tensor5()
self.inputNetwork_Test_Bottom = tensor5()
# Define input shapes to the netwrok
inputSampleShape_Train = (self.batch_Size, 1, self.sampleSize_Train[0], self.sampleSize_Train[1], self.sampleSize_Train[2])
inputSampleShape_Test = (self.batch_Size, 1, self.sampleSize_Test[0], self.sampleSize_Test[1], self.sampleSize_Test[2])
print (" - Shape of input subvolume (Training): {}".format(inputSampleShape_Train))
print (" - Shape of input subvolume (Testing): {}".format(inputSampleShape_Test))
inputSample_Train = self.inputNetwork_Train
inputSample_Test = self.inputNetwork_Test
inputSample_Train_Bottom = self.inputNetwork_Train_Bottom
inputSample_Test_Bottom = self.inputNetwork_Test_Bottom
# TODO change cnnLayers name by networkLayers
self.generateNetworkLayers(cnnLayers,
kernel_Shapes,
self.pooling_scales,
inputSampleShape_Train,
inputSampleShape_Test,
inputSample_Train,
inputSample_Train_Bottom,
inputSample_Test,
inputSample_Test_Bottom,
intermediate_ConnectedLayers)