Do you ever make high-dimensional (multivariate) models?
As my practical experience, I often ask questions when I talked about "I have never worked on a theme that handles image data" and "I have experience on a theme that handles factory sensor data". ..
As a bonus when it comes to this story, the person who asked the question said, "Since the sensor data is one-dimensional data, this person is doing a lot of analysis of one-dimensional data. That said, it's not a big deal compared to image data because it only has multiple sensors and only increases in several dimensions. This person has only done simple data analysis. " It may end up.
It's up to each person to decide whether or not they think it's "easy" in the end, but before making that decision, there are some things that the person who asked the question should know. I haven't understood it, but it is as follows.
When dealing with factory sensor data, the main data may be just one thermometer data. Of course, this data can be analyzed as one-dimensional data, but that may not be enough. In the first place, those who usually work on the data at the factory know that it can be understood as one-dimensional data. The reason why the author consults is that "analysis of one-dimensional data does not solve the problem".
The details are in the analysis of sensor data, but the original sensor data is the "primary data" on this site. It needs to be converted to "secondary data". This conversion uses knowledge of factory batch production methods and discrete production methods.
In secondary data, it is usually two-dimensional or more. Therefore, even the data of one thermometer will be treated as high-dimensional data, but in order to understand the necessity of this conversion, the production method is also involved, so it is quite difficult. I don't understand.
When it comes to creating secondary data, people who know deep learning often interpret it as "such work can be automated using deep learning."
Factory data analysis cannot proceed after data analysis unless it can be explained by the principles of manufacturing. I can't sympathize with the people at the factory, and I can't decide how to operate it.
With that in mind, creating secondary data is not easy, at least for me.
In the field of deep learning, there is talk of seeking explanation, and research is progressing, but it is possible to go into the area of ??manufacturing principles and use it as a technology to manage what is happening in the field now. It looks like it's still a long way off.
Details can be found on the page on the relationship between the purpose and method of data analysis and how to proceed with data science, but in factory data analysis, it is often not important to create a model.
On the other hand, those who ask questions often imagine creating a model and incorporating it into a factory system as a post-data analysis task. Moreover, it is often thought that complicated mathematical models such as deep learning are required.
Modeling is often not important, but the focus is on modeling, and as a result, the conclusion is that "this person has only done simple data analysis." There are so many. .. ..