Software engineers should spend more time towards R&D for better future. It could be healthcare or any other stream. As the need grows so does the software applications, data and infrastructure. I see more and more time and effort is spent on maintenance, handling operational cases. With facility of Cloud Infrastructure, datascience and techniques like HTM its best time to automate these processes. Thereby allowing engineers to spend more quality time thinking for nextgen.

What it does

Predicts hard drive failures using TemporalAnomaly Model

How I built it

Data Ingestion is done using collectors which read SMART data attributes from clusters
SMART Attributes are read at frequency of 2 hours and sent to common database
Adapters feed data from this common datastore to Model (good and bad) to get anomaly score
Anomaly score is used to classify data
Finally the classification is analyzed against actual class

Challenges I ran into

Getting right feature vector
Understanding temporal nature of Harddrive data
Choosing right modeling technique
Performing binary classification

Accomplishments that I'm proud of

I always wanted to start and attempt but couldn't make time for same. Finally through this hackathon i am working on my favorite theory of Machine Intelligence and applying it to solve real problem i see today.

What I learned

Lots of it, Nupic modeling techniques, how important it is to understand data and fetch right feature vector. Python.

What's next for htm-drivefailures

I am presenting this at my work and will continue my efforts with help of community to further test it with real time data and take it from gamma to prod state in future.

Built With

Share this project: