def draw(self,inputNames = None, outputNames = None):
ml.draw(self.Vs, self.W, inputNames, outputNames)
def draw(self,inputNames = None, outputNames = None):
ml.draw(self.Vs+[self.W], inputNames, outputNames)
def draw(self,inputNames = None, outputNames = None):
ml.draw(self.Vs+[self.W], inputNames, outputNames)
def draw(self,inputNames = None, outputNames = None):
weightMatrices = [layer.W for layer in self.layers]
ml.draw(weightMatrices, inputNames, outputNames)
def draw(self, inputNames = None, outputNames = None):
ml.draw(self.Vs, self.W, inputNames, outputNames)
plt.legend(('Train Target','Train NN Output'),loc='lower right',
prop={'size':9})
plt.subplot(nPlotRows,1,3)
plt.plot(Xtest,Ttest)
plt.plot(Xtest,Ytest)
plt.legend(('Test Target','Test NN Output'),loc='lower right',
prop={'size':9})
colors = ('blue','green','red','black','cyan','orange')
for i in range(len(nnet.nhs)):
plt.subplot(nPlotRows,1,i+4)
plt.plot(X,Z[i]) #,color=colors[i])
plt.ylabel('Hidden Unit Outputs')
plt.text(8,0, 'Layer {}'.format(i+1))
plt.subplot(nPlotRows,1,nPlotRows)
ml.draw(nnet.Vs,nnet.W,['x'],['sine'])
plt.draw()
# Now train multiple nets to compare error for different numbers of hidden layers
def experiment(hs,nReps,nIter,X,T,Xtest,Ytest):
results = []
for i in range(nReps):
nnet = NeuralNetwork(1,hs,1)
nnet.train(X,T,weightPrecision=0,errorPrecision=0,nIterations=nIter)
# print "SCG stopped after",nnet.getNumberOfIterations(),"iterations:",nnet.reason
(Y,Z) = nnet.use(X, allOutputs=True)
Ytest = nnet.use(Xtest)
rmsetrain = np.sqrt(np.mean((Y-T)**2))
rmsetest = np.sqrt(np.mean((Ytest-Ttest)**2))
results.append([rmsetrain,rmsetest])