joshlk / cycle_prediction

Predicting end of log time

Predicting Remaining Cycle Time from Ongoing Case

Predicting the remaining cycle time of ongoing cases is one important use case of predictive process monitoring. It is machine learning approach based on survival analysis that can learn from complete/ongoing traces.
we train a neural network to predict the probability density function of the remaining cycle time of a running case.

Documentation:

https://fazaki.github.io/cycle_prediction/

Getting started:

A) pip installation

1. Cd to home dir

cd ~

2. Initialize a virtualenv that uses the Python 3.7 available at home directory

virtualenv -p ~/python-3.7/bin/python3 PROJECTNAME

3. Activate the virtualenv

Windows:

source ~/PROJECTNAME/Scripts/activate

Linux:

source ~/PROJECTNAME/bin/activate

4. Install below packages

pip install cycle-prediction

5. Create a new kernel with the same project name

pip install -U pip ipykernel
ipython kernel install --user --name=PROJECTNAME

6. Use the example notebook

B) Source code installation:

1. Cd to home dir

cd ~

2. Initialize a virtualenv that uses the Python 3.7 available at home directory

Virtualenv -p ~/python-3.7/bin/python3 PROJECTNAME

3. Activate the virtualenv

Windows:

source ~/PROJECTNAME/Scripts/activate

Linux:

source ~/PROJECTNAME/bin/activate

4. Install ipykernel

pip install -U pip ipykernel

5. Clone the repo

git clone https://github.com/fazaki/time-to-event/tree/master
cd time-to-event

6. Install required dependencies:

pip install -e .

7. Use the example notebook

Theory

• Paper publication in progress

References

• https://arxiv.org/abs/1612.02130
• https://ragulpr.github.io/assets/draft_master_thesis_martinsson_egil_wtte_rnn_2016.pdf
• https://verenich.github.io/ProcessSequencePrediction/
• https://github.com/ragulpr/wtte-rnn