Machine learning and intensive reading “Advent Calendar”: vol

nimda November 30, 2025

0 9 5 minutes read

Machine learning and intensive reading “Advent Calendar”: vol

it is very easy to train any model. And the training process is always done in a way that seems to be the same. So we get used to this idea that training any model is the same and easy.

With attol, grid search, and gen ai, “training” machine learning models can be done with a simple “Prompt”.

But the truth is, when we make a model.Fit, behind each model, the process can be very different. And each model itself works differently with the data.

We can see two different trends, almost in two different ways:

On the other hand, we train, use, manipulate, and predict with models (such as generative models) that are more complex.
On the other hand, we are not always able to explain simple models (such as direct regression, direct classification), and intuitive Classimier), and the results of manual iteration.

It is important to understand the models we use. And the best way to understand them is to use them ourselves. Some people do it with Python, r, or other programming languages. But there is still an obstacle for those who do not plan. And these days, understanding AI is important for everyone. In addition, using a programming language can hide certain functions behind existing functions. And it is not defined visually, which means that each operation is not shown properly, because the operation is coded and run, to give results.

So the best test tool, in my opinion, is successful. With formulas that clearly show all the calculation steps.

In fact, when we receive a dataset, most non-cultists will open Excel to understand its contents. This is very common in the business world.

Even many data scientists, myself included, use Excel to take a quick look. And when it's time to explain the results, showing them directly in Excel is often the most effective way, especially in front of management.

In Excel, everything is it's clear. There is no “black box”. You can see every formula, every number, every calculation.

This helps a lot to understand how the models actually work, without shortcuts.

Also, you don't need to install anything. Just a spreadsheet.

I am going to publish a series of articles that address one way or the other – Listen and fill in Machine learning and deep learning models in It's in the mind.

For the “Advent” calendar, I will publish one article per day.

Produced by Gemini: “Advent Calendar” for AI

Who is this series?

For readers, I think these articles provide a realistic perspective. It is to see the feeling of complex formulas.

For developers of ML or AI, WHO, Sometimes, they have never studied the idea – but now, without complex algebra, it is possible, or mathematics, you can open a black box behind the model.Fit. Because in every model, you make a model.Fit. But in reality, the models can be very different.

This is for managers who may not have all the technical background, but which will provide all the visual ideas behind the models. Therefore, combined with the technology of your business, you can better judge if machine learning is really necessary, and which model is right for you.

So, in summary, it is a better understanding of models, training of models, interpretation of models, and links between different models.

Structure of articles

From a practitioner's point of view, we often divide models into the following two categories: supervised learning and unsupported learning.

Then with supervised learning, we have recovery and differentiation. And with unsupported learning, we have a reduction in overlap and overlap.

An overview of machine learning models from a value perspective – image by Author

But surely you have already noticed that some algorithms can share the same or similar, like kn classifier vs regressifier vs regressifier vs regressifier vs regressifier vs regressifier vs regressifier vs regressifier vs regressifier vs regressifier vs regressifier vs regressifier vs regressifier vs descon de regreesos, redress vs vs. “

The Retrieval Tree and the specific retrieving have the same purpose, that is, to perform the retrieving function. But when you try to use them in Excel, you will see that the recursion tree is very close to the classification tree. And the linear regression is close to the neural network.

And sometimes people confuse K-NN with K-Thos. Some may say that their goals are completely different, and that confusing them is the first mistake. But, we also have to admit that we share the same method of calculating distances between data points. So there is a relationship between them.

The same goes for the isolation forest, as we can see that in the random forest where it is also a “forest”.

So I will organize all the models from a theoretical point of view. There are three main methods, and obviously we will see how these methods are used very differently in Excel.

This overview will help us go through all the different models, and connect the dots between many of them.

An overview of machine learning models organized by cross-sectional methods – image by Author

For food-based models, we will calculate local or global distances, between new observations and training data.
In tree-based models, we must define poles or rules that will be used to create feature classes.
For mathematical functions, the idea is to add weights to the elements. Along with training the model, Gradient Forcent is mainly used.
Deep learning models, we will make that the main point is about feature engineering, to create an adequate representation of the data.

In each model, we will try to answer these questions.

Common questions about the model:

What kind of model?
How is the model trained?
What is the hyperpaspambeter of the model?
Can the same model approach be used for regression, isolation, or integration?

Features there are Model:

How are classified features handled?
How are lost values handled?
For continuous features, does scaling make a difference?
How do we measure the importance of one factor?

How can we qualify importance of features? This question will be discussed again. You may know that packages like lime and shape are very popular, and they are model-agnostic. But the fact is that each model behaves differently, and it is interesting and important to interpret directly with the model.

Relationships between different models

Each model will be in a separate story, but we will discuss links to other models.

We will also discuss the relationship between the different models. Since we really open the “external” black box “, we will also know how to make theoretical improvements in other models.

Kna and LDA (cognitive analysis) are very close. The former uses a local distance, and the latter uses a global distance.
Gradient amplification is similar to gradient descent, only the vector space is different.
Direct reordering is also a thing in the classroom.
Label input can be, of course, used for a category feature, and it can be very useful, it has a lot of power, but you have to choose “labels” wisely.
SVM is very close to linear regression, very close to Ridge regression.
Lasso and SVM Use one common criterion to select features or data points. Do you know what the second s in Lasso chooses?

For each model, we will discuss one point that most traditional studies will miss. I call it an unlearned lesson in machine learning modeling.

Exemplary training vs hyperparameter tuning

In these articles, we will focus only on how the models work and how they are trained. We will not discuss the hyperparameter setting, because the process is the same for every model. We usually use grid search.