跑了一个GitHub上的开源代码：github.com/jiangqn/GCN… ，训练过程出现两个错误。
错误一
报错信息：Traceback (most recent call last): File "..\GCN-GAN-pytorch\test_data.py", line 31, in <module> config = yaml.load(open('config.yml')) TypeError: load() missing 1 required positional argument: 'Loader'
原因分析： 我使用的pyyaml是6.0版本，而在YAML5.1版本后弃用了yaml.load(file)这个用法，因为觉得很不安全，5.1版本之后就修改了需要指定Loader，通过默认加载器（FullLoader）禁止执行任意函数，该load函数也变得更加安全
解决办法： 通过以下三种方法都能解决问题
d1=yaml.load(file,Loader=yaml.FullLoader)
d1=yaml.safe_load(file)
d1 = yaml.load(file, Loader=yaml.CLoader)
方法转载于：TypeError: load() missing 1 required positional argument: ‘Loader‘_one_wangtester的博客-CSDN博客
错误二
报错信息：RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [512, 1]], which is output 0 of AsStridedBackward0, is at version 2; expected version 1 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True).
原因分析： 可能是由于Pytorch的版本不同，内置的BACKWARD的流程发生了变化，我的Pytorch版本是1.13.1, 源代码Pytorch版本不详
报错部分代码：
for epoch in range(config['gan_epoches']):
for i, (data, sample) in enumerate(zip(train_loader, sample_loader)):
# update discriminator
discriminator_optimizer.zero_grad()
generator_optimizer.zero_grad()
in_shots, out_shot = data
in_shots, out_shot = in_shots.cuda(), out_shot.cuda()
predicted_shot = generator(in_shots)
_, sample = sample
sample = sample.cuda()
sample = sample.view(config['batch_size'], -1)
real_logit = discriminator(sample).mean()
fake_logit = discriminator(predicted_shot).mean()
discriminator_loss = -real_logit + fake_logit
discriminator_loss.backward(retain_graph=True)
discriminator_optimizer.step()
for p in discriminator.parameters():
p.data.clamp_(-config['weight_clip'], config['weight_clip'])
# update generator
generator_loss = -fake_logit
generator_loss.backward()
generator_optimizer.step()
out_shot = out_shot.view(config['batch_size'], -1)
mse_loss = mse(predicted_shot, out_shot)
print('[epoch %d] [step %d] [d_loss %.4f] [g_loss %.4f] [mse_loss %.4f]' % (epoch, i,
discriminator_loss.item(), generator_loss.item(), mse_loss.item()))
修改生成网络与判别网络的更新流程，修改后代码为：
for epoch in range(config['gan_epoches']):
for i, (data, sample) in enumerate(zip(train_loader, sample_loader)):
in_shots, out_shot = data
in_shots, out_shot = in_shots.cuda(), out_shot.cuda()
predicted_shot = generator(in_shots)
_, sample = sample
sample = sample.cuda()
sample = sample.view(config['batch_size'], -1)
predicted_shot = predicted_shot.squeeze(0)
# update generator
fake_logit = discriminator(predicted_shot)
generator_loss = -fake_logit.mean()
# fake_logit = discriminator(predicted_shot).mean()
# generator_loss = -fake_logit
generator_optimizer.zero_grad()
generator_loss.backward()
generator_optimizer.step()
# update discriminator
real_logit = discriminator(sample)
fake_logit = discriminator(predicted_shot.detach())
discriminator_loss = -real_logit.mean() + fake_logit.mean()
# real_logit = discriminator(sample).mean()
# discriminator_loss = -real_logit + fake_logit
discriminator_optimizer.zero_grad()
discriminator_loss.backward(retain_graph=True)
discriminator_optimizer.step()
for p in discriminator.parameters():
p.data.clamp_(-config['weight_clip'], config['weight_clip'])
out_shot = out_shot.view(config['batch_size'], -1)
mse_loss = mse(predicted_shot, out_shot)
print('[epoch %d] [step %d] [d_loss %.4f] [g_loss %.4f] [mse_loss %.4f]' % (epoch, i,
discriminator_loss.item(), generator_loss.item(), mse_loss.item()))
最后成功运行
