pymfpatch

A modular, end-to-end pipeline for gap-filling EnergyPlus EPW weather files using high-resolution ERA reanalysis (or other reference series) and a hybrid GRU + XGBoost approach.
It automatically parses, cleans, feature-engineers, imputes missing meteorological variables, and writes out a fully compliant EPW-ready for building simulations.

Features

• Dual-stage imputation
  • GRU for sequential, multi-output regression on variables with temporal patterns (temperature, humidity, wind, radiation...)
  • XGBoost for the remaining scalar variables (visibility, ceiling height, albedo, precipitation...)\
• Flexible configuration of
  • which variables to impute with GRU vs XGB\
  • model hyperparameters (learning rate, depth, batch size, early stopping, ...)\
• EnergyPlus EPW input/output with native header preservation\
• Utility functions for parsing, cleaning, feature-engineering and final EPW formatting

Installation

1. Editable install from local source

# clone & install in editable mode
git clone https://your-repo.org/your-org/pymfpatch.git
cd pymfpatch
pip install -e .

2. Install from PyPI

pip install pymfpatch

3. GPU enabling

For fastest training/inference, run on a CUDA-enabled GPU. By default pip install pymfpatch will pull in the CPU-only torch wheel. If you have Python 3.12 and a compatible GPU, for example, you can upgrade to the GPU build with:

pip install torch==2.7.1+cu128 --index-url https://download.pytorch.org/whl/cu128

Quickstart

from pymfpatch import WeatherImputer
# 1) Instantiate the imputer with your reference (ERA) and target station EPWs:
imputer = WeatherImputer(
    path_to_ref = 'Data/ERA/marignane-era.epw',
    path_to_stn   = 'Data/MF/marignane-mf.epw',
)

# 2) Run the imputation pipeline:
imputer.process()

# 3) Write out a fully-imputed EPW:
imputer.write("marignane-imputed.epw")