📰 Dev.to · YMori

Testing if Bayesian ensemble predictions with individual foreign player projections could predict Yakult and Orix going from last place to champions in 2021.

How adding Bayesian corrections and foreign player projections changed NPB team standings predictions — with accuracy data from 8 years of backtesting.

Notes on adding a Stan Bayesian model to an NPB baseball prediction app — Monte Carlo team simulation, credible intervals in Streamlit, and foreign player proje

I migrated my Japanese baseball (NPB) prediction system to Google BigQuery, trained BQML models with SQL window functions, and deployed the API to Cloud Run — a

Building a real-time maritime monitoring system that quantifies the impact of the 2026 Hormuz crisis using free AIS data, Python, and a Raspberry Pi 5.

A 5-tab Streamlit dashboard analyzing Japan vs Venezuela using pitch-by-pitch MLB Statcast data — no coaching tone, just numbers

I tried to beat the Marcel projection system using MLB Statcast tracking data (exit velocity, barrel rate, Whiff%, Stuff+). Here's what happened: +12.1% improve

I added park factor correction to my Marcel+Stan Bayesian NPB standings model. Win MAE didn't change — but here's why that's expected, and what actually improve

How adding park factor (PF) correction to a Marcel+Stan Bayesian NPB standings simulation affected accuracy across 8 seasons (2018-2025), and the resulting 2026

A follow-up to my Marcel analysis: I applied a Bayesian model (Stan/Ridge correction using K%, BB%, BABIP) to the same 2021 Yakult and Orix question — and found

savant-extras v0.4.2: removed OAA/Outfield Jump/Stuff+ (already in pybaseball). Park factors remain as the unique addition.

How I tried to beat the Marcel projection system for NPB (Japanese pro baseball) using Stan/Ridge regression with K%/BB%/BABIP features, and what I learned abou

I analyzed 40 hitters using Driveline OpenBiomechanics C3D data and found that front leg mechanics — not bat speed — may be the key differentiator in exit veloc

In 2021, both pennant winners came from last place the year before. Using Marcel team projections with roster-verified data, I analyzed why — and found a league

I analyzed 61 pro pitchers using Driveline OpenBiomechanics C3D data and found 5 body mechanics factors that explain a 10+ mph gap in pitch speed among pitchers

I added CI/CD to my NPB player performance prediction system: joblib model artifacts, annual metrics JSON, a FastAPI /metrics endpoint, and an 8-step GitHub Act

Streamlit Community Cloud apps sleep after inactivity. HTTP GET returns 200 with a static HTML shell, but the Python app never starts. Use Playwright to actuall

I analyzed pitching and hitting biomechanics using Driveline OpenBiomechanics C3D data, ezc3d, and MediaPipe. Trunk rotation range showed the strongest correlat

I built a Japanese professional baseball (NPB) player performance prediction system using Marcel projection, LightGBM/XGBoost, custom wOBA/wRC+ calculation, and

baseball-field-viz is a Python library for Statcast spray charts in matplotlib — enabling heatmap overlays that pybaseball's built-in spraychart() doesn't suppo

signate-wandb-sync is a PyPI CLI tool that records SIGNATE competition scores to W&B runs, completing the experiment tracking pipeline alongside signate-deploy.

signate-deploy is a PyPI CLI tool that automates SIGNATE competition setup, data download, and submission via GitHub Actions.

kaggle-wandb-sync is a PyPI CLI tool that syncs W&B offline runs from Kaggle Notebooks automatically via GitHub Actions — even when internet is disabled for com

kaggle-notebook-deploy is a PyPI CLI tool that automates Kaggle Notebook deployment via GitHub Actions, so you never have to use the browser editor again.