Webscikit is a set of tools to run a HTTPServer as a JSON Webservice for scikit-learn predictions. It comes with two examples: boston and boston2
It is work in progress, so bug and feature requests are highly appreciated!
Features:
-
The server can handle multiple models. The models and urls are registered at webscikit.conf .
-
Multiple data-scientist could work locally on their own models, and then later deploy their model to the server.
-
The models can be deployed when the server is online.
-
Each model can save additional metadata needed to transform and predict new data.
-
You can easily start a new project with create_project.py newProjectName
-
In the directory examples/ are examples of different models (boston, boston2 etc.) and also example of requests to the server.
How does it work:
- The model gets fitted by the data scientist, gzip-pickled and then uploaded to the server.
- Http-Clients make POST-requests and send json-files to transform / predict new data and get a Json - response back.
If you wan to run the examples:
-
source export_WEBSCIKITMODELSPATH.sh
-
cd server/
-
./webserver.py
-
cd ../example/requests/
-
./curl_boston.sh
-
./curl_boston2.sh