The following project has been developped for bringing a solution to artificial intelligency problematic.
It is currently running on 2 different projects :
SmartIOT - Intelligency in smart objects
Fakerz - Fake websites detection
The aim is to detect unexceptable things. The whole problem is to define in mathematics words how an event can be considered as unexceptable.
[optional] Dockerfile in base project can be used in order to facilitate environment installation.
Docker-compose can be used for having multiple container instances (1 Db container / 2 python container) thanks to docker-compose.yml.dist.
Please note two files are available in 'docker' folder according to your server
rpi/DockerfileRaspberry Debian OS configured thanks to Hypriot team See Hypriot Blogstd/DockerfileStandard configuration (Linux, Mac OS, Windows)
Project is running on Python v2.7
Following librairies are required for compiling project
- PIP for project lib dependencies doc here
- Numpy for using complex mathematics concepts doc here
- SQLAlchemy for managing entities (ORM) doc here
- EVE-SQLAlchemy for easily exposing database through Rest services doc here
The architecture is based on easy concepts :
Subena IA has to detect
unexceptableevents according to calculated stats.
An event at timestamp T contains N values where N is the number of criteria.
Criterion << 1 : N >> Value
Criteria can be grouped by families.
Criterion << N : 1 >> Family
Criteria can be linked with each other.
Criterion << N : N >> Criterion
Thus, used database sub_ia is :
| Tables | Description |
|---|---|
alerts |
Unexceptable events are stored here |
criterion |
Algorithm is based on evaluating values of these |
dependency |
Algorithm calculates dependencies between criteria |
family |
Criteria can be grouped by family (calculation are not impacted) |
settings |
Contains any additional parameters |
stats |
Algorithm calculated relevant correlation between sensors and store value here |
value |
Values for criteria. One line equals to one value for one criterion at a given timestamp |
Thanks to SQLAlchemy, every kind of database should be supported but only MySQL has been effectively used and tested for now.
An init MySQL script can be found at the following path scripts/init-schema.sql
An example of database with data can be found at scripts/example.sql. In this example, values are get from 8 criteria corresponding to 8 sensors.
Hearth of algorithm is contained in package folder.
alerts folder contains methods for storing unexceptable events
learning parameter and structure learning algorithm
model defines classes linked with database
For launching project just execute main script :
python scripts/main.py -v
Author : Jessy HANZO