Today I practiced Linear Regression in the Machine Learning Zoomcamp. I learned how to split a dataset into training, validation, and test sets using NumPy.
#mlzoomcamp #MachineLearning

This week I continued with the ML Zoomcamp 2026.

The last week were about 2. ML for Regression.
- Explore+prepare a real dataset
- Selecting useful features
- Implement linear reg. w/ & w/o regularization
- Validate+compare reg. models with RMSE

➡️Next step: 3. ML for Classification

#mlzoomcamp

7/ Introducing categorical variables can create numerical instability if duplicate columns make the feature matrix uninvertible. Regularization solves this by adding a small parameter to the matrix's diagonal, bringing extreme weights back under control. #mlzoomcamp

6/ Visualizing predictions on a chart isn't an objective way to measure a model's performance. Root Mean Squared Error (RMSE) provides a definitive metric, squaring the differences to penalize larger errors and prevent positive and negative errors from canceling each other out. #mlzoomcamp

5/ Training a linear regression model isn't magic; it relies on the Normal Equation. Because real-world feature matrices aren't square and cannot be inverted directly, we use the Gram matrix to calculate the closest approximate solution for the weights. #mlzoomcamp

4/ A linear regression prediction starts with a bias term, representing the baseline estimate when absolutely nothing else is known about the observation. Each known feature then adjusts that baseline via its specific weight. #mlzoomcamp

3/ Splitting data sequentially is dangerous if the dataset contains inherent ordering. Shuffling the data with a fixed random seed before splitting ensures a clean, reproducible distribution across the training, validation, and test sets. #mlzoomcamp

2/ Real-world price data usually features a long-tail distribution, which tends to confuse machine learning models. Applying a logarithmic transformation removes this tail, reshaping the data into a model-friendly normal distribution. #mlzoomcamp

Kicking off Module 2 of #mlzoomcamp. The focus shifts entirely to a hands-on regression project, taking raw dataset characteristics and building a linear regression model from scratch to predict car prices.

Last September 14th I started the wonderful ML Zoomcamp 2026.

These 2 weeks were about 1. Introduction to ML:
- Difference between ML and rule-based systems
- What supervised ML is
- CRISP-DM
- The model selection step
- Setting up the environment

➡️Next step: 2. ML for Regression

#mlzoomcamp

Finished Module 1 of ML Zoomcamp 2026! 🎉 ML vs rule-based, CRISP-DM, & NumPy/Pandas.
✨ Takeaway: ML is a lifecycle—business & data prep matter most.
🤔 Best part: Shifting from coding rules to letting data find them via vectorization!
➡️ Next: Regression!
#mlzoomcamp @alexeygrigorev.bsky.social

7/ Working with randomly generated arrays requires reproducibility. Fixing the random seed ensures pseudorandom algorithms produce the exact same sequence of numbers across different environments, standardizing your outputs. #mlzoomcamp

6/ Model selection carries the multiple comparisons problem: evaluate enough models on one validation set, and one looks good by pure chance. A separate test set acts as an essential audit against this statistical luck. #mlzoomcamp

I just completed Module 1: Introduction to Machine Learning from ML Zoomcamp 2026🎉

This module covered interesting part and start with basic Machine learning, rule based system, numpy and pandas:

Find my Colab playbook

Link: colab.research.google.com/drive/1CUrim...

#mlzoomcamp @Alexey Grigorev

5/ The most valuable question in the first phase of CRISP-DM is whether machine learning is actually necessary. If a simple rule-based system works, building a full ML pipeline is unnecessary overhead. #mlzoomcamp

4/ Supervised machine learning is fundamentally an applied mathematics problem. Training is simply a search for a function that maps an input feature matrix to a known target variable. #mlzoomcamp

3/ Hard-coding rules for complex systems becomes a maintenance nightmare. Machine learning solves this by flipping the roles: the desired outcome becomes the input, allowing the algorithm to automatically extract the rules. #mlzoomcamp

2/ Experts price assets by applying patterns internalized from past examples. Machine learning replicates this mechanism by feeding features and target prices into an algorithm to automatically extract those patterns. #mlzoomcamp

Kicking off Module 1 of #mlzoomcamp. The curriculum frames machine learning as just one part of a much larger engineering process spanning from initial business understanding straight through to deployment.