https://github.com/aliyahscoding/weather-insights

End-to-end weather mini-suite: fetch, clean, feature, backtest (naive vs SARIMA vs ML), and Streamlit dashboard. Reproducible, fast, no API keys.
https://github.com/aliyahscoding/weather-insights

data-science forecasting pandas python reproducible-research sarima sklearn streamlit time-series

Last synced: about 2 months ago
JSON representation

End-to-end weather mini-suite: fetch, clean, feature, backtest (naive vs SARIMA vs ML), and Streamlit dashboard. Reproducible, fast, no API keys.

• Host: GitHub
• URL: https://github.com/aliyahscoding/weather-insights
• Owner: aliyahscoding
• Created: 2025-10-02T14:55:06.000Z (9 months ago)
• Default Branch: main
• Last Pushed: 2025-10-05T04:59:47.000Z (9 months ago)
• Last Synced: 2025-10-05T06:18:10.250Z (9 months ago)
• Topics: data-science, forecasting, pandas, python, reproducible-research, sarima, sklearn, streamlit, time-series
• Language: Python
• Homepage:
• Size: 97.7 KB
• Stars: 0
• Watchers: 0
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Weather Insights

A tidy, reproducible pipeline for daily weather: ingest → preprocess → backtest baselines/SARIMA/ML → Streamlit dashboard.

## TL;DR

- **SARIMA** beats seasonal naive by ~X% RMSE on Austin test set (last 365 days).

- **Ridge ML** with lag/rolling/calendar features matches SARIMA within error bars and trains in seconds.

## How to run

python -m venv .venv

.venv\Scripts\activate

pip install -r requirements.txt

python -m src.ingest --location "Austin,US" --start 2018-01-01 --end 2024-12-31 --outfile data/raw/austin_2018_2024.csv

python -m src.preprocess --in data/raw/austin_2018_2024.csv --out data/processed/austin_proc.csv

python -m src.modeling.backtest --dataset data/processed/austin_proc.csv --model sarima --horizon 7

streamlit run app/streamlit_app.py

## Repo layout

weather-insights/

data/{raw,processed,external}

src/{ingest.py,preprocess.py,modeling/,viz.py,utils.py}

app/streamlit_app.py

reports/{figures,metrics}

tests/

.github/workflows/ci.yml

## Results

| model           | MAE | RMSE |

|-----------------|-----|------|

| seasonal_naive  | ... | ...  |

| sarima          | ... | ...  |

| ridge_ml        | ... | ...  |

![Test window](reports/figures/test_pred.png)

## Notes

- Weekly seasonality is strong; SARIMA captures it.

- Simple ML with lags/rolls/calendar nearly ties SARIMA.