扩展PyTorch

本篇文章中包含如何扩展torch.nntorch.autograd和使用C库来编写自定义的C扩展工具。

扩展torch.autograd

添加操作autograd需要Function为每个操作实现一个新的子类。回想一下,Function使用autograd来计算结果和梯度,并对操作历史进行编码。每个新功能都需要您实现两种方法:

  • forward() - 执行操作的代码。如果您指定了默认值,则可以根据需求使用任意参数,其中一些参数可选。这里支持各种Python对象。Variable参数在调用之前会被转换Tensor,并且它们的使用情况将在graph中注册。请注意,此逻辑不会遍历lists/dicts/和其他任何数据的结构,并且只考虑被直接调用的Variables参数。如果有多个输出你可以返回单个TensorTensor格式的元组。另外,请参阅Function文档查找只能被forward()调用的有用方法的说明。

  • backward() - 计算梯度的公式. 它将被赋予与输出一样多的Variable参数, 其中的每一个表示对应梯度的输出. 它应该返回与输入一样多的Variable, 其中的每一个表示都包含其相应输入的梯度. 如果输入不需要计算梯度 (请参阅needs_input_grad属性),或者是非Variable对象,则可返回None类.此外,如果你在forward()方法中有可选的参数,则可以返回比输入更多的梯度,只要它们都是None类型即可.

你可以从下面的代码看到torch.nn模块的Linear函数, 以及注解

# Inherit from Function
class Linear(Function):

    # bias is an optional argument
    def forward(self, input, weight, bias=None):
        self.save_for_backward(input, weight, bias)
        output = input.mm(weight.t())
        if bias is not None:
            output += bias.unsqueeze(0).expand_as(output)
        return output

    # This function has only a single output, so it gets only one gradient
    def backward(self, grad_output):
        # This is a pattern that is very convenient - at the top of backward
        # unpack saved_tensors and initialize all gradients w.r.t. inputs to
        # None. Thanks to the fact that additional trailing Nones are
        # ignored, the return statement is simple even when the function has
        # optional inputs.
        input, weight, bias = self.saved_tensors
        grad_input = grad_weight = grad_bias = None

        # These needs_input_grad checks are optional and there only to
        # improve efficiency. If you want to make your code simpler, you can
        # skip them. Returning gradients for inputs that don't require it is
        # not an error.
        if self.needs_input_grad[0]:
            grad_input = grad_output.mm(weight)
        if self.needs_input_grad[1]:
            grad_weight = grad_output.t().mm(input)
        if bias is not None and self.needs_input_grad[2]:
            grad_bias = grad_output.sum(0).squeeze(0)

        return grad_input, grad_weight, grad_bias

现在,为了更方便使用这些自定义操作,推荐使用apply方法:

linear = LinearFunction.apply

我们下面给出一个由非变量参数进行参数化的函数的例子:

class MulConstant(Function):
    @staticmethod
    def forward(ctx, tensor, constant):
        # ctx is a context object that can be used to stash information
        # for backward computation
        ctx.constant = constant
        return tensor * constant

    @staticmethod
    def backward(ctx, grad_output):
        # We return as many input gradients as there were arguments.
        # Gradients of non-Tensor arguments to forward must be None.
        return grad_output * ctx.constant, None

你可能想检测你刚刚实现的backward方法是否正确的计算了梯度。你可以使用小的有限差分法(Finite Difference)进行数值估计。

from torch.autograd import gradcheck

# gradcheck takes a tuple of tensors as input, check if your gradient
# evaluated with these tensors are close enough to numerical
# approximations and returns True if they all verify this condition.
input = (Variable(torch.randn(20,20).double(), requires_grad=True), Variable(torch.randn(30,20).double(), requires_grad=True),)
test = gradcheck(Linear.apply, input, eps=1e-6, atol=1e-4)
print(test)

扩展 torch.nn

nn模块包含两种接口 - modules和他们的功能版本。你可以用两种方法扩展它,但是我们建议,在扩展layer的时候使用modules, 因为modules保存着参数和buffer。如果使用无参数操作的话,那么建议使用激活函数,池化等函数。

添加操作的功能版本已经在上面的章节中已经介绍了。

增加一个Module

由于nn大量使用autograd。所以, 添加一个新的Module类需要实现一个Function类, 它会执行对应的操作并且计算梯度。我们只需要很少的代码就可以实现上面Linear模块的功能。现在,我们需要实现两个函数:

  • __init__ (optional) - 接收kernel sizes内核大小,特征数量等参数,并初始化parameters参数和buffers缓冲区。
  • forward() - 实例化Function并使用它来执行操作。它与上面显示的functional wrapper非常相似。

下面是实现Linear模块的方式:

class Linear(nn.Module):
    def __init__(self, input_features, output_features, bias=True):
        super(Linear, self).__init__()
        self.input_features = input_features
        self.output_features = output_features

        # nn.Parameter is a special kind of Variable, that will get
        # automatically registered as Module's parameter once it's assigned
        # as an attribute. Parameters and buffers need to be registered, or
        # they won't appear in .parameters() (doesn't apply to buffers), and
        # won't be converted when e.g. .cuda() is called. You can use
        # .register_buffer() to register buffers.
        # nn.Parameters require gradients by default.
        self.weight = nn.Parameter(torch.Tensor(output_features, input_features))
        if bias:
            self.bias = nn.Parameter(torch.Tensor(output_features))
        else:
            # You should always register all possible parameters, but the
            # optional ones can be None if you want.
            self.register_parameter('bias', None)

        # Not a very smart way to initialize weights
        self.weight.data.uniform_(-0.1, 0.1)
        if bias is not None:
            self.bias.data.uniform_(-0.1, 0.1)

    def forward(self, input):
        # See the autograd section for explanation of what happens here.
        return LinearFunction.apply(input, self.weight, self.bias)

    def extra_repr(self):
        # (Optional)Set the extra information about this module. You can test
        # it by printing an object of this class.
        return 'in_features={}, out_features={}, bias={}'.format(
            self.in_features, self.out_features, self.bias is not None
        )

编写自定义的C扩展

即将发布。不过现在你可以在GitHub上找到一些例子 。

译者署名

用户名 头像 职能 签名
Song 翻译 人生总要追求点什么