mummify 0.3.3

Git + Logging for ML

mummify

Install

pip install mummify

About

"You suck at Git. And logging. But it's not your fault."

mummify is git+logging for people who do machine learning.

mummify is simple, opinionated, and built directly on top of git.

mummify comes packaged with three main functions:

• log to automatically log and commit
• history to view the commit graph tree
• switch to switch to a different commit

Quick Start

Import mummify at the top of your machine learning script (in this case, model.py) and add mummify.log(<message>) at the very end of your script:

import mummify
from sklearn.datasets import load_wine
from sklearn.model_selection import train_test_split
from sklearn.neighbors import KNeighborsClassifier

data = load_wine()
y = data.target
X = data.data

X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, random_state=42)

model = KNeighborsClassifier()
model.fit(X_train, y_train)
accuracy = model.score(X_test, y_test)
mummify.log(f'Test accuracy: {accuracy:.3f}')

When you call python model.py from the command line mummify will initialize a specialized .mummify git directory, create a mummify.log file, and keep track of model performance:

Whenever you make a change to your model, think: swap in a new algorithm:

import mummify
from sklearn.datasets import load_wine
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression

data = load_wine()
y = data.target
X = data.data

X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, random_state=42)

model = LogisticRegression()
model.fit(X_train, y_train)
accuracy = model.score(X_test, y_test)
mummify.log(f'Test accuracy: {accuracy:.3f}')

And re-run python model.py mummify will update the mummify.log file and save the state of your model:

To view the mummify log history from the command line you can run mummify history:

max-mbp:quick-start max$ mummify history

*  HEAD mummify-f0e66a82
*  mummify-75dea5e9
*  mummify-start

max-mbp:quick-start max$

And to rewind history (go back to a previous state), just grab the mummify identifier that you want to switch to from the mummify.log file and run mummify switch <id> from the command line:

mummify will preserve all state history on a switch and keep the mummify.log file immutable.