Encoding For Predictive Process Monitoring (enc4ppm)

enc4ppm is a Python package than provides common process mining encodings.

Documentation and reference.

Installation

Using pip:

pip install enc4ppm

Example

The following example performs frequency encoding with latest payload for next activity prediction task:

import pandas as pd

from enc4ppm.frequency_encoder import FrequencyEncoder
from enc4ppm.constants import LabelingType

# Load log
log = pd.read_csv('bpic2012.csv')

# Create encoder
encoder = FrequencyEncoder(
    labeling_type=LabelingType.NEXT_ACTIVITY,
    include_latest_payload=True,
    attributes=['AMOUNT_REQ'],
)

# Encode log
encoded_log = encoder.encode(log)

Features

• Frequency, simple-index and complex-index encodings
• Next activity, remaining time and outcome labelings
• Save encoder to disk for later use
• Freeze encoder on training set, then use it on unseen data (automatic handling of unknown values)
• Standardize numerical features
• Convert categorical features to one-hot encoding, or keep them as strings
• Add time features (time since case start and time since last event) to the encoding

Development

Documentation

Documentation is provided by mkdocs. To build and push the documentation website to GitHub, run the following command: mkdocs gh-deploy.

License

MIT License