rorolite¶
rorolite is an open-source command-line tool to deploy Machine Learning applications to your own server. It provides simple interface to provision the server to install all the required dependencies and deploy the ML application as an API.
This is a lite version of the rorodata platform.
System Requirements¶
The target server should be running Ubuntu 16.04.
How to use¶
Write a rorolite.yml specifying the host ipaddress and the services.
runtime: python3
# IP address/hostname of the target server
host: 1.2.3.4
# username on the target server
user: alice
services:
# run the predict function in credit_risk_service module as an API on port 8000
- name: api
function: credit_risk_service.predict
port: 8000
# run gunicorn process port 8080
- name: webapp
command: gunicorn webapp:app -b 0.0.0.0:8080
port: 8080
Either a function or a command can be specified as a service. When a function is specified as a service, rorolite used the firefly to deploy it as a service.
The server needs to provisioned once to install all the necessary system software and base dependencies specified by the runtime mentioned in the rorolite.yml file. All the application dependencies are installed on every deploy.
The currently available runtimes are:
- python3
- python3-keras
To provision the server, run:
$ rorolite provision
...
To deploy your project, run:
$ rorolite deploy
Deploying project version 7...
...
Services are live at:
api -- http://1.2.3.4:8000/
webapp -- http://1.2.3.4:8080/
The deploy command pushes your code to the server, sets up a virtual env, installs all the dependencies from your requirements.txt file and starts the specified services.
Inspect the running services using the ps command.
$ rorolite ps
...
api RUNNING pid 23796, uptime 0:02:07
The logs command allows inspecting logs of any service.
$ rorolite logs api
2017-10-25 04:13:12 firefly [INFO] Starting Firefly...
2017-10-25 04:15:12 predict function called
The run command allows running any command on the remote server.
$ rorolite run python train.py
starting the training...
reading the input files...
building the model...
saving the model...
done.
Or you can even start a jupyter notebook server.
$ rorolite run:notebook
...
Copy/paste this URL into your browser when you connect for the first time,
to login with a token:
http://1.2.3.4:8888/?token=7f53b445100a5edc0d035fb7ce53061ff7dae351a107ebd4
Copying files to/from remote server can be done using put/get commands. A directory /volumes/data is created during provisioning for storing data files, models etc.
$ rorolite put data/loans.csv /volumes/data/
...
[1.2.3.4] put: data/loans.csv -> /volumes/data/loans.csv
$ rorolite get /volumes/data/model.pkl models/model.pkl
...
[1.2.3.4] download: models/model.pkl <- /volumes/data/model.pkl