There are many different types of data, so there are many jobs that use data science and that data science is useful for.
By the way, there are many explanations in the world that "what to do in data science = create and use models". "Creating a model" means calculating the coefficients of a model, such as regression analysis or deep learning, using data.
If you think that "creating and using models" is data science, your work in data science will be limited. However, it's hard to see what data science work is like when models is not important.
On this page, we're organizing our data science work in contrast to when models is important versus when they don't.
To think about what data science work looks like, it seems better to think about when models is important and when they don't. The following is the main theory after explaining the case.
If the model is important, it seems better to use the model from the beginning and if it is not.
If you're starting to decide how to use a model, for example, you'll want to use the techniques of "recognizing what's in the image" or "choosing the one that suits the user's preferences" from the start.
If you already have some software and know-how to use these technologies, you can devise inputs and outputs to make them functions for apps and homepages, robots, and so on.
This is the case when you know about different models and use that knowledge to start something new or solve problems.
It's like making a model from the beginning by going through the specific formulas of various models.
When a model is not important, it is a job to examine and think about the meaning and content of the data.
As for what is data, we think about the history, background, accuracy, etc. of the data. Think about things that aren't data. "To verify this fact, how do you take this data, how do you transform it?" When you have a choice and you get lost, you take this kind of data and look at it like this.
Even When model is not important, you may try to create a model, but by applying the model, it is a way to check the contents of the data. We'll see if the model is accurate or low, but it's important to know if it's high or low, and it is not important if it's not high.
If you can create a model with high accuracy, you may want to think that "I did it!
Causal inference is an area of investigation that explores the relationship between cause and effect. There are many models of causal reasoning in the world, but considering the time when investigating the causes of what is actually happening in practice, I think it is better to think that causal reasoning is "when the model is not important".
When model is not important, it seems to be good to say whether it is emergency or not.
When immediate response is necessary for "accident occurrence", "abnormal occurrence", etc.
The first response is on the spot, focusing on experience, and it may be a good thing to perform deeper data analysis at the next stage of it. In order to reduce the spread of damage even a little, hurry up.
There are projects in the world that seem to be unresolved even if it feels like the place and the eternal theme of the work, and it is thought that "it would be good if it disappeared".
In the factory, there are occurrences of defective products of unknown cause and machine failure.
If the model is important, it often doesn't seem to be that steep. Even if the earlier it is, the better, it seems that it is often okay to proceed with a guide of about a few months to a year.
The sooner an emergency case is effective, and the later it is, the more likely it is that it makes no sense to do. It's like "now," "today," or "at most within a week." In these cases, the sooner you get results, the greater the effect.
Non-emergency cases can last for a period of several months. In that case, even if it can take several months, if it can be solved, the effect will be large.
When models are important, they are often imaged as the next stage of efforts to improve the efficiency of the company's operations by introducing IT such as accounting systems and human resources systems. "Next to it system introduction, we will introduce artificial intelligence (AI) system".
If the model is important, there are many way to proceed.
In case of emergency, you may not be able to report like a report, and you may want to create a graph in excel screen to complete it. Anyway, since earlyness is often important, it is not possible to proceed with "launch a project and ..."
If it is not emergency, it is better to use Problem-solving steps.
The pages on this site, Data Analysis by Excel, Data Analysis by R, and Data Analysis by Python, are summarized based on the know-how I used and what I wanted to do when the model was not important.
It seems that labor costs are often very high because of the long time and the large number of people involved. In addition, if the story proceeds to create a system or equipment in the end, the cost will also be required.
The time is short, and in some cases, even one person can do it, so labor costs are small. In addition, even if you do something as a countermeasure, you may be able to improve it without spending money.
If the model is important, it is in the area of the consultancy or the data analysis company.
On the other hand, When model is not important, it is difficult for such a company to do it because of the tight delivery time and the depth of the content.
If a model is not important, I think the person responsible for data science is the person who is responsible for the problem or the problem, or who can think about things in a position close to it.
When you apply a model to a company or society and find a good case, it becomes a job when modeling is important. For example, if there is something like "I'm doing work that people see, let's make this possible with image recognition AI", it will become a job.
On the other hand, regardless of whether the model applies or not, there are many things in the company and in society that it is good that the facts are in the form of data or that the facts are unders been unders been unders been made through the data. There are many different types of data, so there are many things you can do with data.
So when you compare the amount of work that a model is important to the amount of work that a model is not, I think the latter is overwhelmingly more.
However, the latter seems to be not established as a work of data science.
For example, if you find a cause from your data to solve a problem, you will notice that the cause may be this, and the success criterion is whether to take action and resolve the problem. If you use these success criteria, even if your data is not a direct reference to cause and effect, it can lead to problem solving.
I have experience, but regardless of what kind of data you have, it seems that work often ends up stalling if you proceed in a way that seeks a model that fits the data perfectly anyway.
NEXT Mathematical Science in Data ScienceTweet