Which model to choose?

Question

PHP

If I smoke a little, I can figure out the php code, I understand what it is and how to work with github, mysql ... I have long wanted to try to make my own neuron. There is a practical task and actually I would like to do it:

There is a number that needs to be predicted, along with the number there is a number of values, incl. text.
For example such data:

• location,
  
• number of people
  
• senior group
  
• grade
  

*perhaps even the day of the week and the time of day, in short +/- such inputs
There are several hundred thousand lines for training...

Based on the known data, it is necessary to predict what the final grade will be.

I can’t understand if I can implement something like this on my server with php-ml, if there is a suitable model? Other languages ​​are probably not worth it, unless python but then the most primitive example...

Answer 1

you can. in general, any neuron is primarily an equation with a large number of parameters.
your task is very similar to regression analysis
here is a link that helps explain what a neural network is (machine learning) https://qna.habr.com/q/565845

Answer 2

There is no point in programming this task. Decide what you will use - ns or something else.
Then use ready-made packages. Of the neural networks, the restricted Boltzmann machine is best suited.

Answer 3

If your score is some ranked set of values ​​(well, like at school: 1, 2, 3, 4 ... 12 or like at a university - F, E, D, C, B, A or like a scale of epidemic danger "green" - "yellow" - "orange" - "red") and / or at least one of your signs - a non-continuous numerical value (location, group leader, day of the week, etc.) - then the choice of model is absolutely obvious and unambiguous. This is a classification.
Is it necessary (is it reasonable) - to implement it independently? Well, only if you really want to train, you have a lot of free time and little depends on the quality of the result. Because in any self-respecting package that claims to be an ML tool, there is an implementation of at least one of the classification algorithms. Including the one you are referring to (php-ml). Another thing is its quality - but here it’s even too lazy to check, because in the same Python / Scikit-Learn there are more than a dozen implemented classification algorithms (logistic regression, k-nearest neighbors, support vectors, naive bayes classifier, decision tree, random forest, AdaBoost, gradient boosting - maybe I missed something) whose quality has been tested a thousand times, performance has been proven. Choose, try look for the most appropriate for your data. And to use their knowledge of Python you need - at the lowest level.
Yes, among neural networks there are networks adapted to solving this class of problems. But in general, it is advisable to use them (neural networks) not instead of well-functioning MD algorithms, but where it is problematic to use such algorithms in their pure form. And in the simplest cases, like yours (the number of parameters is about 10, all of them are clearly highlighted, the data (can be) summarized in a table), using a neural network is like using a Kamaz to transport a mother-in-law to a dacha - it is possible, but to put it mildly, it is not advisable and extremely uncomfortable.