jshdmm/weather-predict-berlin

Weather Prediction in Berlin

A LightGBM regressor predicting hourly temperature (°C) trained daily from Open-Meteo data. Part of an MLOps pipeline — full architecture, training/serving code, and API at github.com/jshdmm/weather-predict. Importantly, feature engineering and model performance optimization were not the focus of this project, but rather hosting a baseline model using CI/CD with GitHub Actions, automated testing, orchestration using a cronjob, and containerization with Docker. Additionally, the model can be accessed through FastAPI endpoints.

Files

• model.pkl — the trained model (joblib)
• results.json — training metadata + test MAE
• plot.png — predicted vs. actual on the held-out test period

Usage

from huggingface_hub import hf_hub_download
import joblib

model = joblib.load(hf_hub_download("jshdmm/weather-predict-berlin", "model.pkl"))

Retrained automatically once per day using a cronjob — results.json always test MAE, not a fixed benchmark.