def evaluate(cvae, testLoader, exDir, e=1, classifier=None): #e is the epoch
cvae.eval()
#Load test data
xTest, yTest = prep_data(iter(testLoader).next(), cvae.useCUDA)
print 'saving a set of samples'
if cvae.useCUDA:
z = Variable(torch.randn(xTest.size(0), opts.nz).cuda())
else:
z = Variable(torch.randn(xTest.size(0), opts.nz))
ySmile = Variable(torch.Tensor(yTest.size()).fill_(1)).type_as(yTest)
samples = cvae.decode(ySmile, z).cpu()
save_image(samples.data, join(exDir, 'smile_epoch' + str(e) + '.png'))
yNoSmile = Variable(torch.Tensor(yTest.size()).fill_(0)).type_as(yTest)
samples = cvae.decode(yNoSmile, z).cpu()
save_image(samples.data, join(exDir, 'no_smile_epoch' + str(e) + '.png'))
#check reconstructions after each 10 epochs
outputs, outMu, outLogVar, outY = cvae(xTest)
bceLossTest, klLossTest = cvae.loss(rec_x=outputs,
x=xTest,
mu=outMu,
logVar=outLogVar)
predLabel = torch.floor(outY)
classScoreTest = binary_class_score(predLabel, yTest, thresh=0.5)
print 'classification test:', classScoreTest.data[0]
save_image(xTest.data, join(exDir, 'input.png'))
save_image(outputs.data, join(exDir, 'output_' + str(e) + '.png'))
rec1, rec0 = label_switch(xTest.data, yTest, cvae, exDir=exDir)
# for further eval
if e == 'evalMode' and classer is not None:
classer.eval()
yPred0 = classer(rec0)
y0 = Variable(torch.LongTensor(yPred0.size()).fill_(0)).type_as(yTest)
class0 = binary_class_score(yPred0, y0, thresh=0.5)
yPred1 = classer(rec1)
y1 = Variable(torch.LongTensor(yPred1.size()).fill_(1)).type_as(yTest)
class1 = binary_class_score(yPred1, y1, thresh=0.5)
f = open(join(exDir, 'eval.txt'), 'w')
f.write('Test MSE:' + str(F.mse_loss(outputs, xTest).data[0]))
f.write('Class0:' + str(class0.data[0]))
f.write('Class1:' + str(class1.data[0]))
f.close()
return (bceLossTest).data[0] / xTest.size(0), classScoreTest.data[0]
#check reconstructions after each 10 epochs
outputs, outMu, outLogVar, outY = cvae(Variable(xTest))
bceLossTest, klLossTest = vae_loss_fn(rec_x=outputs.data,
x=xTest,
mu=outMu,
logVar=outLogVar)
maxVal, predLabel = torch.max(outY, 1)
classScoreTest = torch.eq(predLabel,
yTest).float().sum() / yTest.size(0)
print 'classification test:', classScoreTest.data[0]
save_image(xTest, join(exDir, 'input.png'))
save_image(outputs.data, join(exDir, 'output_' + str(e) + '.png'))
label_switch(xTest, yTest, cvae, exDir=exDir)
cvae.save_params(exDir=exDir)
losses['total'].append(epochLoss / Ns)
losses['kl'].append(epochLoss_kl / Ns)
losses['bce'].append(epochLoss_bce / Ns)
losses['test_bce'].append(
(bceLossTest).data[0] / xTest.size(0)) #append every epoch
losses['dis'].append(epochLoss_dis / Ns)
losses['gen'].append(epochLoss_gen / Ns)
losses['class'].append(epochLoss_class / Ns)
losses['test_class'].append(classScoreTest.data[0])
if e > 1:
plot_losses(losses, exDir, epochs=e + 1)
