← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6577.a · 生成模型:GPT-6

分组编码要保留各自类别轴

理解等级:中级

本例将多个因子分别one-hot编码后拼接,总长度为1+3+6+40+32+32=114。解码时必须按每组cardinality拆分,再各自取样或argmax;把114维整体当一个类别分布,会丢失多个因子的组合含义。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      CARDINALITY = [1, 3, 6, 40, 32, 32] #1 class EncoderCausalFactors(nn.Module): #2 def __init__(self, image_dim, factor_dim): super(EncoderCausalFactors, self).__init__() self.image_dim = image_dim self.factor_dim = factor_dim hidden_dim = 1000 #3 self.fc1 = nn.Linear(image_dim, hidden_dim) #4 self.fc2 = nn.Linear(hidden_dim, hidden_dim) #4 self.fc3 = nn.Linear(hidden_dim, factor_dim) #4 self.softplus = nn.Softplus() #4 self.sigmoid = nn.Sigmoid() #5 def forward(self, img): img = img.reshape(-1, self.image_dim) #6 hidden1 = self.softplus(self.fc1(img)) #7 hidden2 =

    参考来源

    - 位置:《Causal AI》第09章《通用反事实推断算法》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Causal AI/_epub-src/OEBPS/Text/chapter-9.xhtml) - 结论依据:摘录中关于分组编码要保留各自类别轴的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:CARDINALITY = [1, 3, 6, 40, 32, 32] #1 class EncoderCausalFactors(nn.Module): #2 def __init__(self, image_dim, factor_dim): super(EncoderCausalFactors, self).__init__() self.image_dim = image_dim self.factor_dim = factor_dim hidden_dim = 1000 #3 self.fc1 = nn.Linear(image_dim, hidden_dim) #4 self.fc2 = nn.Linear(hidden_dim, hidden_dim) #4 self.fc3 = nn.Linear(hidden_dim, factor_dim) #4 self.softplus = nn.Softplus() #4 self.sigmoid = nn.Sigmoid() #5 def forward(self, img): img = img.reshape(-1, self.image_dim) #6 hidden1 = self.softplus(self.fc1(img)) #7 hidden2 =