MIDAS(2)

Work in progress implementation of MIDAS in pytorch.

Please see the CHANGELOG below for updates on the development of this package. Please also note that while preliminary testing has been conducted on this implementation, it is still in the early stages of development and we cannot guarantee its performance. Documentation for the main model and methods can be found as docstrings in the MIDAS2.py script.

In addition to migrating to torch, this new version adds the following functionality:

• Models can be fit on X and used to impute on new data X'
• Automatic detection of column-types

Example usage

The major syntactical difference to MIDASpy is that MIDAS2 follows the sklearn API, with fit and transform methods of an imputer object.

from MIDAS2 import model as md

# Create a MIDAS model
mod = md.MIDAS()

# Fit the model to data
mod.fit(X, epochs = 10)

# Multiply impute missing data
X_imputed = mod.transform(m = 10)

CHANGELOG

• Alpha release including combination rules function (26/06/2025)
• Renamed the main module from 'MIDAS2' to 'model' (19/12/2024)
• Restructured the package for easier install (19/12/2024)