Comparison between rule engine and ML model - xlaszlo

“ Rules based systems ” Usually as ML A good starting point for the project .

  

The focus of data science is to ask questions

The job of a data scientist is to answer questions through statistical tools . Sometimes some answers can be expressed in the form of models , But it's not necessary .

Data scientists focus first on problems, not answers . This requires them to build a framework （ Use code and data ）, In order to verify subsequent answers .

 

 

Building a rule - based system is simple ？

Establishing rules sounds simple , isn't it? ？

1. First from " Price >1000 dollar , It is true , Otherwise it is false " Such content begins ;
2. Then you add logical operators , Such as AND、OR and NOT;
3. then , You adjust the constant , demonstrate ;
4. Then you add loops and branches to simplify the logical expression .

Is this a rule-based system ？ It seems very simple , But in the end , You may want to delegate the project to a software engineering team ; They are better at writing code , Then the data scientists pass DSL Enter business rules .

later , You will find the rule engine +DSL It doesn't satisfy you ....

  

Is it difficult to model artificial intelligence ？

Suppose you prepare and build a framework that validates the model ：

It is really difficult to implement a new machine learning model from scratch . But the biggest recent advance in this field is ： There are already a number of high-quality models that are easy to use .

You can learn from spacy、huggingface or imagenet Wait until the pre trained model starts to solve the standard task .

You have structured data *BOOST Model , Use sklearn Of LogReg There is no problem with building a baseline model .

In a rapidly developing field like machine learning , You not only need to update your skills , You also need to update your beliefs about what you accept as a reasonable first attempt .