PyTorch
    PyTorch模型源代码中的ctx参数，总结如下：

    • ctx是context的缩写，翻译成”上下文; 环境”
    • 主要用在自定义torch.autograd.Function中
    • ctx专门用在静态方法中
    • self指的是实例对象; 而ctx用在静态方法中，调用的时候不需要实例化对象，直接通过类名就可以调用，所以self在静态方法中没有意义
    • 自定义的forward()方法和backward()方法的第一个参数必须是ctx; ctx可以保存forward()中的变量，以便在backward()中继续使用，下一条是具体的示例
    • **ctx.save_for_backward(a, b)**能够保存**forward()**静态方法中的张量，从而可以在backward()静态方法中调用，具体地，下面地代码通过**a, b = ctx.saved_tensors**重新得到a和b

    • **ctx.needs_input_grad**是一个元组，元素是True或者False，表示**forward()**中对应的输入是否需要求导，比如ctx.needs_input_grad[0]指的是下面forwad()代码中indices是否需要求导

      1. class SpecialSpmmFunction(torch.autograd.Function):
      2.   """
      3.   Special function for only sparse region backpropataion layer.
      4.   """
      5.   # 自定义前向传播过程
      6.   @staticmethod
      7.   def forward(ctx, indices, values, shape, b):
      8.       assert indices.requires_grad == False
      9.       a = torch.sparse_coo_tensor(indices, values, shape)
      10.       ctx.save_for_backward(a, b)
      11.       ctx.N = shape[0]
      12.       return torch.matmul(a, b)
      14.   # 自定义反向传播过程
      15.   @staticmethod
      16.   def backward(ctx, grad_output):
      17.       a, b = ctx.saved_tensors
      18.       grad_values = grad_b = None
      19.       if ctx.needs_input_grad[1]:
      20.           grad_a_dense = grad_output.matmul(b.t())
      21.           edge_idx = a._indices()[0, :] * ctx.N + a._indices()[1, :]
      22.           grad_values = grad_a_dense.view(-1)[edge_idx]
      24.       if ctx.needs_input_grad[3]:
      25.           grad_b = a.t().matmul(grad_output)
      26.       return None, grad_values, None, grad_b

      ctx还能调用很多方法

      1. class _FunctionBase(object):
      2.   # no doc
      3.   @classmethod
      4.   def apply(cls, *args, **kwargs): # real signature unknown
      5.       pass
      7.   def register_hook(self, *args, **kwargs): # real signature unknown
      8.       pass
      10.   def _do_backward(self, *args, **kwargs): # real signature unknown
      11.       pass
      13.   def _do_forward(self, *args, **kwargs): # real signature unknown
      14.       pass
      16.   def _register_hook_dict(self, *args, **kwargs): # real signature unknown
      17.       pass
      19.   def __init__(self, *args, **kwargs): # real signature unknown
      20.       pass
      22.   @staticmethod # known case of __new__
      23.   def __new__(*args, **kwargs): # real signature unknown
      24.       """ Create and return a new object.  See help(type) for accurate signature. """
      25.       pass
      27.   dirty_tensors = property(lambda self: object(), lambda self, v: None, lambda self: None)  # default
      29.   metadata = property(lambda self: object(), lambda self, v: None, lambda self: None)  # default
      31.   needs_input_grad = property(lambda self: object(), lambda self, v: None, lambda self: None)  # default
      33.   next_functions = property(lambda self: object(), lambda self, v: None, lambda self: None)  # default
      35.   non_differentiable = property(lambda self: object(), lambda self, v: None, lambda self: None)  # default
      37.   requires_grad = property(lambda self: object(), lambda self, v: None, lambda self: None)  # default
      39.   saved_tensors = property(lambda self: object(), lambda self, v: None, lambda self: None)  # default
      41.   saved_variables = property(lambda self: object(), lambda self, v: None, lambda self: None)  # default
      43.   to_save = property(lambda self: object(), lambda self, v: None, lambda self: None)  # default