Some machine learning projects produce great results for the client. Some don't, for a variety of reasons. What makes the difference?
How can you manage scope, schedule, and costs for a machine learning project?
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
Very open question- here are my two cents:
Is the problem solvable? Some problems (like predicting long term stock markets or, for the UK police, finding the difference between pornographic imagery and dunes link) are just very difficult.
Quality of data. If your data is an unreliable pile of garbage this can make your life a living nightmare of data cleaning, imputation and tons of exceptions.
How willing is the client to actually accept all data findings. I have seen this time over time. Some people are really good at cherry picking the one indicator in their preferred direction while the rest of them points in the other direction.
Usability of the result. If you produce a result that does not scale well (like a computational expensive predictor or similar) it might just not be feasible to use it.
