时间序列数据伤亡的自动编码器
我正在尝试找出两次系列向量之间的链接。
例如 ： x =温度变化一年； y =应变测量一年；
我是机器学习的新手。 我不知道要使用哪种无监督算法来识别可能的因果关系并了解两个向量之间的联系？ 我可以使用自动编码器吗？
I am trying to figure out the link between 2-time series vectors.
for example :
X = temperature variation for one year;
Y = strain measurement for one year;
I am new to machine learning.
I cannot figure out which unsupervised algorithm to use for identifying possible causality and understanding the link between the 2 vectors?
Can I use autoencoders?
如果你对这篇内容有疑问，欢迎到本站社区发帖提问 参与讨论，获取更多帮助，或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱，如果其他用户或者作者回复了您的评论，将不能在第一时间通知您！
发布评论
评论（1）
通常，复杂的机器学习方法不是因果关系的核心，通常不是必需的。大多数情况下，棘手的部分是使模型估算正确的数量。对于因果关系的入门，这是一个可访问的资源。
自动编码器通常很有趣，因为它们可以学习复杂数据的较低维度表示。这在因果关系中很少是一个实际问题，当然也不是两个变量。
要解决因果问题，首先您需要指定您的问题：您是否对变量之间的因果效应感兴趣？您知道哪一个是原因，哪一个效果？然后，您将需要考虑需要哪些数据来回答问题。您是否有观察和介入（例如随机对照试验）数据？是时间序列吗？然后您需要表达因果效应，例如平均治疗效果使用机器学习来估计它。
In general, complex machine learning methods are not at the core of causality and often not necessary. Most of the time the tricky part is in getting models to estimate the correct quantity. For a primer on causality, this is an accessible resource.
Autoencoders are typically interesting because they can learn lower-dimensional representations of complex data. This is rarely a practical problem in causality, and certainly not for two variables.
To tackle a causal problem, first you need to specify your question: Are you interested in the strength of a causal effect between the variables? Do you know which one is the cause, and which one the effect? Then you will need to think about what data you need to answer the question. Do you have observational and interventional (e.g. randomized control trial) data? Is it a time series? Then you need to express you causal effect, e.g. the average treatment effect, and only then can you use machine learning to estimate it.