Machine Learning Practice
Stratified Sampling Preserves Labels, Not Every Important Difference
Stratification can preserve class proportions in a train-test split. It does not automatically balance every feature or prevent data leakage.
AI · VISION · EVERYDAY SYSTEMS
Journal
Machine Learning Practice
Macro and Micro Averages Can Tell Different Stories About an AI Classifier
Macro and micro averaging answer different questions about classifier performance. Use a small example to see why class balance matters.
AI Evaluation
A Random Seed Helps Repeat an AI Experiment, but Does Not Guarantee Identical Results
Record the environment and sources of randomness alongside a seed when comparing machine-learning experiments.
Language Model Basics
A Model’s Context Window Also Has to Leave Room for Its Answer
A model’s context window includes more than your document. Budget for instructions, history, tools, and the answer you want it to produce.
AI Evaluation
A Confusion Matrix Shows Which Labels an AI Model Mixes Up
Read a confusion matrix by checking its axes and label order, then use a small worked example to find errors hidden by an overall accuracy figure.
Applied Machine Learning
Changing a Classification Threshold Changes Which AI Mistakes You See
See how a classifier’s decision threshold changes false alarms and missed cases, using a small worked example and a fixed evaluation set.
AI System Design
Where You Split a Document Can Change an AI Search Answer
Document chunk boundaries can separate an answer from its condition. Test retrieval with real questions before choosing a chunk size for an AI system.
AI Concepts
Reading Embeddings Without Inventing Names for Every Dimension
Understand an embedding as a learned numerical representation without assuming each coordinate has a tidy human-readable meaning.
AI Fundamentals
Training and Inference Put an AI Model to Different Kinds of Work
Distinguish training from inference so questions about AI cost, data use and model changes are directed at the right process.