Data Scientist: Why is my cutting edge model still not in production?
Scenario 1:
IT Team: We are still implementing the scoring engine in Go/Java/C++.
Scenario 2:
IT Team: We are still figuring out how to read the model.pkl file you provided.
. .
and the push to production pang continues ..
- Now available in PyPI -
pip install titus2
View the complete changelog here.
This is a subset-fork of the original repository which has been completely migrated to Python 3.
The Portable Format for Analytics (PFA) is a specification for scoring/inference engines: event-based processors that perform predictive or analytic calculations. It is a model interchange format which helps smoothen the transition from statistical model development to large-scale and/or online production.
Titus (Python 2) (API) was originally Open Data's complete implementation of PFA for Python. It can be used for model development as well as to execute the scoring engine. Titus 2 is a fork of Titus which is actively being maintained.
Titus 2 uses a number of open source projects to work properly:
- avro-python3
- numpy
- pytz
- pyyaml
- ply
The above packages are available via pip
and are automatically installed during setup.
Titus requires Python 3.4+ to run. It can be installed via pip/pip3 as follows:
$ pip install titus2
or you can directly install the latest build from github repository via
$ pip install git+https://github.com/animator/titus2.git
After installation please run the following elementary example in python
from titus.genpy import PFAEngine
pfa = {"input": "double",
"output": "double",
"action": [
{"+": ["input", 100]}
]}
engine, = PFAEngine.fromJson(pfa)
l = [1.0, 2.0, 3.0, 4.0, 5.0]
for num in l:
print(num, engine.action(num))
See the Hadrian wiki for user guide and tutorials.
Please raise an issue here.
Want to contribute? Great!
Please raise an issue and send a pull request.
- Write MORE Tests for
scripts/*
. - Add
scikit-learn
model export to PFA tutorials.