azimgd / logistic-regression-prediction

CTR Prediction

##install and demo

• cat redis.model | redis-cli
• npm install
• node lib/index

##test curl -X POST -d '{"deviceExtBrowser": "Firefox", "bannerExtSize": "300x250", "deviceLanguage": "de", "deviceExtType": "tablet" }' -H "Content-Type: application/json" http://localhost:3000

##todo

• Validate mathematical operations using high precision calculators (maybe wolfram)
• Remove hard 16 units precision format in logistic service
• Accept input request even if it's not proper json (bodyParser.json() doesn't accept requests without text/json header)
• Input request validation to controller methods
• Write more integration + unit tests on controllers
• Fix redis null hget

##commands

• npm run build to build project into lib folder
• npm run dev to start auto build on save

##sidenote Logistic tests are a bit over-engineered from my opinion, that was only to show how i do mocking/spying etc. Current values that are being tested against are not having high precision, i had to use external library (https://github.com/MikeMcl/big.js) to make mathematical operations due to the nature of JS with floating numbs. In any case math operators are wrapped in separate functions so they could be swapped easily. I had not much time to write (integration ?!) tests for controller methods, but in my opinion tests for calculateCtr method and dataSelectors are cruicial.