, it is rather easy to coach any model. And the training process is at all times done with the seemingly same method fit. So we get used to this concept that training any model is comparable and straightforward.
With autoML, Grid search, and Gen AI, “training” machine learning models could be done with a straightforward “prompt”.
But the truth is that, once we do model.fit, behind each model, the method could be very different. And every model itself works very in a different way with the info.
We are able to observe two very different trends, almost in two opposite directions:
- On the one hand, we train, use, manipulate, and predict with models (resembling generative models) increasingly complex.
- Alternatively, we should not at all times able to explaining easy models (resembling linear regression, linear discriminant classifier), and recalculating results by hand.
It is crucial to know the models we use. And the most effective option to understand them is to implement them ourselves. Some people do it with Python, R, or other programming languages. But there remains to be a barrier for individuals who don’t program. And nowadays, understanding AI is important for everybody. Furthermore, using a programming language can even hide some operations behind already existing functions. And it isn’t visually explained, meaning that every operation isn’t clearly shown, because the function is coded then run, to only give the outcomes.
So the most effective tool to explore, in my view, is Excel. With the formulas that clearly show every step of the calculations.
The truth is, once we receive a dataset, most non-programmers will open it in Excel to know what’s inside. This could be very common within the business world.
Even many data scientists, myself included, use Excel to take a fast look. And when it’s time to elucidate the outcomes, showing them directly in Excel is usually probably the most effective way, especially in front of executives.
In Excel, every little thing is visible. There is no such thing as a “black box”. You’ll be able to see every formula, every number, every calculation.
This helps quite a bit to know how the models really work, without shortcuts.
Also, you do not want to put in anything. Only a spreadsheet.
I’ll publish a series of articles about methods to understand and implement machine learning and deep learning models in Excel.
For the “Advent Calendar”, I’ll publish one article per day.
Who is that this series for?
For college students who’re studying, I feel that these articles offer a practical standpoint. It’s to make sense of complex formulas.
For ML or AI developers, who, sometimes, haven’t studied theory — but now, without complicated algebra, probability, or statistics, you’ll be able to open the black box behind model.fit. Because for all models, you do model.fit. But in point of fact, the models could be very different.
This can be for managers who may not have all of the technical background, but to whom Excel will give all of the intuitive ideas behind the models. Subsequently, combined with your corporation expertise, you’ll be able to higher judge if machine learning is basically mandatory, and which model is perhaps more suitable.
So, in summary, It’s to higher understand the models, the training of the models, the interpretability of the models, and the links between different models.
Structure of the articles
From a practitioner’s standpoint, we normally categorize the models in the next two categories: supervised learning and unsupervised learning.
Then for supervised learning, we now have regression and classification. And for unsupervised learning, we now have clustering and dimensionality reduction.
But you surely already notice that some algorithms may share the identical or similar approach, resembling KNN classifier vs. KNN regressor, decision tree classifier vs. decision tree regressor, linear regression vs. “linear classifier”.
A regression tree and linear regression have the identical objective, that’s, to do a regression task. But once you attempt to implement them in Excel, you will note that the regression tree could be very near the classification tree. And linear regression is closer to a neural network.
And sometimes people confuse K-NN with K-means. Some may argue that their goals are completely different, and that confusing them is a beginner’s mistake. BUT, we also must admit that they share the identical approach of calculating distances between the info points. So there may be a relationship between them.
The identical goes for isolation forest, as we are able to see that in random forest there is also a “forest”.
So I’ll organize all of the models from a theoretical standpoint. There are three essential approaches, and we are going to clearly see how these approaches are implemented in a really different way in Excel.
This overview will help us to navigate through all the various models, and connect the dots between lots of them.
- For distance-based models, we are going to calculate local or global distances, between a brand new statement and the training dataset.
- For tree based models, we now have to define the splits or rules that shall be used to make categories of the features.
- For math functions, the concept is to use weights to features. And to coach the model, the gradient descent is especially used.
- For deep learning models, we are going to that the essential point is about feature engineering, to create adequate representation of the info.
For every model, we are going to try to reply these questions.
General questions on the model:
- What’s the character of the model?
- How is the model trained?
- What are the hyperparameters of the model?
- How can the identical model approach be used for regression, classification, and even clustering?
How features are modelled:
- How are categorical features handled?
- How are missing values managed?
- For continuous features, does scaling make a difference?
- How will we measure the importance of 1 feature?
How can we qualify the importance of the features? This query may even be discussed. It’s possible you’ll know that packages like LIME and SHAP are very fashionable, and so they are model-agnostic. But the reality is that every model behaves quite in a different way, and it is usually interesting, and vital to interpret directly with the model.
Relationships between different models
Each model shall be in a separate article, but we are going to discuss the links with other models.
We may even discuss the relationships between different models. Since we truly open each “black box”, we may even know methods to make theoretical improvement to some models.
- KNN and LDA (Linear Discriminant Evaluation) are very close. The primary uses a neighborhood distance, and the latter uses a world distance.
- Gradient boosting is identical as gradient descent, only the vector space is different.
- Linear regression can be a classifier.
- Label encoding could be, type of, used for categorical feature, and it may be very useful, very powerful, but you will have to decide on the “labels” correctly.
- SVM could be very near linear regression, even closer to ridge regression.
- LASSO and SVM use one similar principle to pick features or data points. Do you recognize that the second S in LASSO is for selection?
For every model, we also will discuss one particular point that almost all traditional courses will miss. I call it the untaught lesson of the machine learning model.
List of articles
Below there shall be an inventory, which I’ll update by publishing one article per day, starting December 1st!
See you very soon!
…
