lossG.backward()
optimizer_G.step()
# ====================
# Save to tensorborad
# ====================
if (i == 0 or (i % opt.n_display_step == 0)):
board_train.add_scalar('Generater/loss_G', lossG.item(),
iterations)
board_train.add_scalar('Discriminator/loss_D', lossD.item(),
iterations)
board_train.add_scalar('Discriminator/loss_D_real',
lossD_real.item(), iterations)
board_train.add_scalar('Discriminator/loss_D_fake',
lossD_fake.item(), iterations)
board_add_image(board_train, 'fake_image', gen_imgs, iterations)
# Monitor trainnig progresses
print("epoch={}, iters={}, loss_G={:.5f}, loss_C={:.5f}".format(
epoch, iterations, lossG, lossD))
# ============
# Save images
# ============
generator.eval()
with torch.no_grad():
fake_imgs = generator(z_fixed)
save_image(fake_imgs[0],
os.path.join(opt.dir_out, opt.exper_name) +
"/fake_image_epoches{}_batch0.png".format(epoch),
normalize=True)
# 勾配計算
loss_G.backward()
# backward() で計算した勾配を元に、設定した optimizer に従って、重みを更新
optimizer_G.step()
#====================================================
# 学習過程の表示
#====================================================
if( step == 0 or ( step % args.n_display_step == 0 ) ):
board_train.add_scalar('Generater/loss_G', loss_G.item(), iterations)
board_train.add_scalar('Discriminator/loss_D', loss_D.item(), iterations)
board_train.add_scalar('Discriminator/loss_D_real', loss_D_real.item(), iterations)
board_train.add_scalar('Discriminator/loss_D_fake', loss_D_fake.item(), iterations)
board_add_image(board_train, 'fake image', G_z, iterations)
print( "epoch={}, iters={}, loss_G={:.5f}, loss_D={:.5f}".format(epoch, iterations, loss_G, loss_D) )
#====================================================
# test loss の表示
#====================================================
if( step == 0 or ( step % args.n_display_test_step == 0 ) ):
model_G.eval()
model_D.eval()
n_test_loop = 0
test_iterations = 0
loss_D_total = 0
loss_D_real_total = 0
loss_D_fake_total = 0
loss_G_total = 0
# 勾配計算
loss.backward()
# backward() で計算した勾配を元に、設定した optimizer に従って、重みを更新
optimizer.step()
#====================================================
# 学習過程の表示
#====================================================
if (step == 0 or (step % args.n_display_step == 0)):
board_train.add_scalar('Model/loss', loss.item(), step + 1)
print("step={}, loss={:.5f}".format(step + 1, loss))
board_add_image(board_train,
'input image',
image,
step + 1,
n_max_images=4)
#board_train.add_histogram('output[0]', output[0], step+1)
#print( "step={}, targets=({}), output=({:.5f},{:.5f})".format(step+1, targets[0].item(), output[0,0], output[0,1]) )
#----------------------------------------------------
# 正解率を計算する。（バッチデータ）
#----------------------------------------------------
# 確率値が最大のラベル 0~9 を予想ラベルとする。
# dim = 1 ⇒ 列方向で最大値をとる
# Returns : (Tensor, LongTensor)
_, predicts = torch.max(output.data, dim=1)
if (args.debug and n_print > 0):
print("predicts.shape :", predicts.shape)