For non-technical readers: This page explains every technical term used across the documentation in plain English. Bookmark this page as a quick reference.
Machine Learning & AI
| Term | Plain English | Where Used |
|---|---|---|
| Machine Learning (ML) | Software that learns patterns from historical data to make predictions about the future | Everywhere |
| MLOps | The practice of managing ML models like software products — versioning, deploying, and monitoring them in production | Everywhere |
| CatBoost | A prediction algorithm (“model”) that excels at learning from mixed data types — numbers and categories together | Training |
| Regression | A type of prediction that outputs a number (e.g., “142 bikes will be rented”) rather than a category | Training |
| Hyperparameters | Settings that control how the model learns — like tuning dials on a radio to get the clearest signal | Parameters |
| Bayesian Optimization | A smart method for finding the best hyperparameters by learning from previous results rather than testing randomly | Training |
| RMSE | Root Mean Square Error — a measure of how far predictions are from actual values. Lower is better | Training |
| SHAP | A method that explains why the model made a specific prediction by showing which factors mattered most | Explainability |
Data & Features
| Term | Plain English | Where Used |
|---|---|---|
| Feature | A piece of information the model uses to make predictions (e.g., temperature, hour of day, humidity) | Feature Engineering |
| Feature Engineering | The process of creating new data columns from existing ones to help the model learn better | Feature Engineering |
| Lag Features | Values from previous time periods — e.g., “how many bikes were rented 1 hour ago?” — that help predict future demand | Feature Engineering |
| Data Pipeline | An automated sequence of steps that collects, validates, cleans, and prepares data for the model | Data Pipeline |
| Data Catalog | A registry listing every dataset in the system, its format, and where it’s stored | Data Catalog |
| Pandera | A library that enforces data quality rules — rejecting bad data before it reaches the model | Data Pipeline |
| Data Drift | When real-world data stops matching what the model was trained on, causing predictions to degrade | Data Quality |
Infrastructure & Deployment
| Term | Plain English | Where Used |
|---|---|---|
| Docker | Technology that packages software into standardized containers that run identically everywhere | Infrastructure |
| Docker Compose | A tool for running multiple Docker containers (services) together as one system | Infrastructure |
| Container | A lightweight, isolated package containing everything needed to run a piece of software | Infrastructure |
| API | Application Programming Interface — a way for software to request predictions or data from the system | API Reference |
| FastAPI | The Python framework powering the prediction API — known for speed and automatic documentation | Inference |
| Endpoint | A specific URL where the API accepts requests (e.g., /predict for generating forecasts) | API Reference |
| Rate Limiting | Restricting how many API requests a user can make per minute to prevent abuse | Inference |
| Argon2id | An industry-standard password hashing algorithm used to securely verify API keys | Inference |
Orchestration & Automation
| Term | Plain English | Where Used |
|---|---|---|
| Airflow | A scheduling platform that runs automated workflows on a timetable, like a smart cron job | Orchestration |
| DAG | Directed Acyclic Graph — a visual representation of a workflow showing which steps depend on others | Orchestration |
| Kedro | A Python framework for building reproducible data science pipelines with clear structure | Data Pipeline |
| Optuna | A tool that automatically searches for the best model settings using Bayesian optimization | Training |
| Automated Retraining | The system’s ability to detect performance degradation and retrain the model without human intervention | Retraining |
Monitoring & Observability
| Term | Plain English | Where Used |
|---|---|---|
| Prometheus | A system that collects and stores metrics (numbers) about system health — CPU, memory, request counts | Operations |
| Grafana | A dashboard tool that visualizes Prometheus metrics with charts, graphs, and alerts | Operations |
| MLflow | A platform for tracking experiments, versioning models, and managing the model lifecycle | Experiments |
| MinIO | An S3-compatible object storage system used to store model files and data artifacts | Infrastructure |
| Evidently | A monitoring library that detects data drift using statistical tests (like the KS-test) | Retraining |
| KS-test | Kolmogorov-Smirnov test — a statistical method that measures if two data distributions are different | Data Quality |
Metrics & Evaluation
| Term | Plain English | Where Used |
|---|---|---|
| MAE | Mean Absolute Error — average difference between predicted and actual values. Easier to interpret than RMSE | Training |
| R² Score | How well the model explains the variation in data. 1.0 = perfect, 0.0 = no better than guessing the average | Training |
| PSI | Population Stability Index — measures how much the data distribution has shifted over time | Operations |
| A/B Testing | Splitting traffic between two models to compare which performs better in production | Inference |
Related Pages
Problem Statement
Business context and objectives behind the forecasting system.
System Overview
High-level architecture showing how all components connect.