You're navigating the world of algorithms. How do you blend innovation with your trusted framework?

Change is the new constant. It important to understand and learn new concepts which do help in bringing a new fresh perspective and helps resolve problems efficiently and effectively. There are n number of ways to solve a problem and so are the algorithms available. Say for example you have sorting algorithms available, start with bubble sort going through with Quick sort. Remember, learn and adapt. See if the new algorithm you came to know about is a great fit for your problem and is flexible enough to cater other needs.

Assessing compatibility of the new algorithm with the existing framework is quiet Important. We need have the confidense that the algorithm we are going to introduce, fits in well with the framework and also that it's flexible enough to adhere to the future changes. Check that the algorithm goes well with the framework, it's Scalable, it's flexible.

I usually ask my clients for real world CSV data files comprised of sensor data, eg MEMS data. We then develop the concept in Matlab/our own tools. The CSV data files have the big plus of basing the algorithmic concept on real data which can be played back in a deterministic way. I've never been a big fan of simulated data, as it's only good for a few simple spot checks. After achieving satisfactory performance, we then move onto streaming data, as these introduce unexpected scenarios/artefacts into the mix, which usually requires finetuning of the algorithmic parameters.

The new algorithm when have been integrated with the current framework, the existing testcases should be ran to validate that the features work as expected also that the system is in stable condition. Every single testcase should be ran and made sure that no functionality is broken. Make sure you do performance testing, making sure the performance requirements are met and also do make sure to have a check on memory leak. Unwanted resource usage and if algorithm is causing memory problem then that's a big issue again. To mitigate each and every possible problem, make sure you do full scale testing. All the unit testcases along with Integration testing should be covered up

When there's a new algorithm being integrated with the system, make sure you have an iterative approach. Go step by step and make iterative progress. Make sure at each step you have your testcases in place and run them to make sure the functionality is as expected. Have your performance and memory leak test whereever required. This brings in confidence between team members as they can now see the progress and outcome as expected. A need for change and other evaluations can also be considered as we are going ahead with iterative changes. There can be scenarios where a change might be needed, with this approach we can incorporate those changes and make sure the other steps are updated accordingly.