Final project for web technologies course @ UniMoRe.
This webapp tries to address the challenges of in-house stock management for medium-sized teams such as @mmr-driverless.
In particular, the focus of the project is on electronic components, but most of the code would be valid for any kind of discrete item.
Noteworthy features are:
- Protection mechanism: most valuable/scarce components can be protected. Users must require a DL approval before use.
- Naive recommendation system: user-based collaborative filtering
- Real-time dashboard for easy management
- Front-end independent APIs
- DB access synchronization (guaranteed by Django)
- Location uniqueness guaranteed by SQL constraints
Make sure pipenv is installed.
Locally install dependencies, then open virtual-environment shell, with:
pipenv install
pipenv shellIf pipenv complains about an already active venv, solve manually or force it with --anyway option.
Make sure there is no SQLite DB in root folder (where this README is located).
Create migrations (python3 -m manage makemigrations) if model has been changed.
Run the following to setup a mock environment with some users and some components:
bash db_setup.shFeel free to explore its internals to understand what's going on. The expected output is the creation of dev_db.sqlite3 in root folder; the DB is filled in with stock data as mentioned above.
Now you are ready to start the development server:
python3 -m manage runserver 8080Login as admin with password 123 and start to explore.
The recommendation is a user-based collaborative filtering. Each user is characterized by a (presumably sparse) vector of stars.
For each user U, its k-neighbors (K parametrized in analytics.recommendation) are extracted from the high dimensional space. This is the set of users that has an influence on the prediction for U.
The assumption under which this recommender works, is that if users similar to U liked an item i, also U will like it.
Given this assumption, the prediction is easy: stars for each element are averaged. Items that survive a certain filter (thresholding + slicing N-top) will be proposted to the user in the You may want to start from section.
More subtle strategies can be used, like weighting each star with the correlation between U and the user that gave the star; but for small instances, the current method showed to perform well.
Some illustrative test-cases have been written in core (logic-only tests) and analytics (api-oriented tests) applications.
This software is distributed under MIT License.