如何在chainer中检查模型的参数编号

Question

如何检查我的chainer模型的参数编号？它与Keras中的model.count_params()类似吗

Answer 1

如果要查看参数中的值，可以使用

# model.params gives you a generator object which you could iterate and see
for i in model.params():
  print(i)

如果您想查看图层和形状的名称，则可以使用此选项

for i in model.links():
  print(i)

如果尚未初始化输入，则只能检查基于每单位的偏差。像这样

total = 0
for idx, i in enumerate(model.params()):
  if idx%2 != 0: # Skipping the weights and looking at biases
    total += len(i)

print(total)

如果您已经定义了输入，那么权重将不会是None，您也可以通过这样的轻微修改来计算其值

total = 0
for idx, i in enumerate(model.params()):
  if idx%2 != 0:
    total += len(i)
  else:
    total += len(i)*len(i[0]) # len inputs * len units

print(total)

'''
Now for different models, there are different methods so we can use try-except to make it better but be careful as the results might be wrong in some cases. usually, they will be correct. There might be cases when you have nested layers in the model and you have to check recursively.
'''

total = 0
for idx, i in enumerate(model.params()):
  try:
    try:
       total += len(i)
    except:
       total += len(i)*len(i[0])

  except:
       pass