Barclays Interview Questions | Improving Models
Question
When modifying an algorithm, how do you know that your changes are an improvement over not doing anything?
in progress
0
Machine Learning
4 years
3 Answers
683 views
Great Grand Master 0
Answers ( 3 )
We look at the performance metrics like accuracy, precision, F1 score, recall according to the problem we are solving. This can say whether the changes are improving the models performance or not.
Mainly two ways a model can be checked if performance is better
1. Metric basis – Certain metrics that will tell us whether the performance of your model has increased or not like accuracy, recall, F! etc
2. Business rule – In every industry there are some business objective and rules that are decided by stakeholders, if your model is getting close to it, that means you have done a good job!
Test the changes on a hold-out data set and always set seed so that you are able to reproduce the
results. If you are getting better accuracy or a better result on any target metric, your changes are working.