def splitintobatches(self, trainingdata, batchsize): #splits trainingdata into batches of size batchsize
N = len(trainingdata.dic)
if batchsize > N:
batchsize = N
k = int(np.ceil(N/batchsize))
Batch_sets=[0]*k
Batch_sets[0]=TrainingDataSgfPass()
Batch_sets[0].dic = dict(list(trainingdata.dic.items())[:batchsize])
for i in range(k-1):
Batch_sets[i]=TrainingDataSgfPass()
Batch_sets[i].dic=dict(list(trainingdata.dic.items())[i*batchsize:(i+1)*batchsize])
Batch_sets[k-1]=TrainingDataSgfPass()
Batch_sets[k-1].dic = dict(list(trainingdata.dic.items())[(k-1)*batchsize:N])
number_of_batchs = k
return[number_of_batchs, Batch_sets]
def Learnsplit(self, trainingdata, eta, batch_size, stoch_coeff, error_function, trainingrate, error_tolerance, maxepochs):
N = len(trainingdata.dic)
splitindex = int(round(N*trainingrate))
trainingset, testset = TrainingDataSgfPass(), TrainingDataSgfPass()
trainingset.dic = dict(list(trainingdata.dic.items())[:splitindex])
testset.dic = dict(list(trainingdata.dic.items())[splitindex:])
error = [error_tolerance+1]
epochs = 0
while error[-1:][0] > error_tolerance and epochs < maxepochs:
epochs += 1
self.Learn(trainingdata, 1, batch_size, stoch_coeff, error_function)
error.append(self.PropagateSet(testset,error_function))
return [error,epochs]