子组件从url中获取参数_获取模型中所有的参数名和参数值

1. 先来看看没有任何参数的模型

class MyModel(nn.Module):
    def __init__(self):
        print("run __init__")
        super(MyModel, self).__init__()
        
    def get_parameter(self):
        print("run output_parameter")
        print(type(self.named_parameters()))
        # 注意只有当模型具有参数时，self.named_paramters()迭代器才会有值
        parameters_li = [w for w in self.named_parameters()]
        print(parameters_li)

model = MyModel()
model.get_parameter()

输出：

run __init__
run output_parameter
<class 'generator'>
[]

2. 再来看看有参数模型

class MyModel(nn.Module):
    def __init__(self, vocab_size, embedding_dim):
        print("run __init__")
        super(MyModel, self).__init__()
        self.word_embeds = nn.Embedding(vocab_size, embedding_dim)
        
    def output_parameter(self):
        print("run output_parameter")
        print(type(self.named_parameters()))
        # 遍历模型参数
        for name, param in self.named_parameters():
            print(name, param.size())

model = MyModel(10, 3)
model.output_parameter()

输出：

run __init__
run output_parameter
word_embeds.weight tensor([[-0.1012,  0.4560, -0.6860],
        [-0.7001, -0.2842,  0.8513],
        [ 1.5048, -0.6218, -0.5249],
        [-0.4236, -1.5744,  1.5675],
        [-0.3904,  0.0451, -1.6845],
        [ 0.4153, -0.0883,  1.1259],
        [ 0.6372,  0.5849,  0.6626],
        [-0.0312, -0.0170,  0.2092],
        [-2.1321, -0.7156, -1.7835],
        [-0.5531,  0.4732, -0.6747]])

3. 可以直接操作模型参数

class MyModel(nn.Module):
    def __init__(self, vocab_size, embedding_dim):
        print("run __init__")
        super(MyModel, self).__init__()
        self.word_embeds = nn.Embedding(vocab_size, embedding_dim)
        
    def output_parameter(self):
        print("run output_parameter")
        # 遍历模型
        for name, param in self.named_parameters():
            print(name, param.data)
        # 直接操作参数
        print(self.word_embeds.weight)

model = MyModel(10, 3)
model.output_parameter()

输出：

run __init__
run output_parameter
word_embeds.weight tensor([[ 0.3675,  0.3408,  0.1545],
        [ 0.7712,  0.5315, -0.9827],
        [-0.0351,  1.9727,  0.8911],
        [-0.2503,  0.6004, -0.6274],
        [ 0.7397, -0.0736,  0.3195],
        [-0.1690,  0.5512, -0.0648],
        [-0.8347, -0.9638,  0.5015],
        [-1.0703,  0.6631,  1.4699],
        [ 0.9917, -1.1515,  0.3213],
        [ 0.5136, -0.1434,  0.3430]])
Parameter containing:
tensor([[ 0.3675,  0.3408,  0.1545],
        [ 0.7712,  0.5315, -0.9827],
        [-0.0351,  1.9727,  0.8911],
        [-0.2503,  0.6004, -0.6274],
        [ 0.7397, -0.0736,  0.3195],
        [-0.1690,  0.5512, -0.0648],
        [-0.8347, -0.9638,  0.5015],
        [-1.0703,  0.6631,  1.4699],
        [ 0.9917, -1.1515,  0.3213],
        [ 0.5136, -0.1434,  0.3430]], requires_grad=True)

再一次重复，嵌入层参数就是一个词向量矩阵，每行表示一个词，每列表示维度。

嗯，按照我老师的教诲，一时的努力不是真努力，要细水长流。