Suppose, you've done all you need to do to build a prediction model. Despite your hard work and high expectation, the prediction model you've designed performs poorly.
What to do next then?
After having a discussion with a good friend, he suggested me to review Andrew Ng's advices for debugging predictive analytic (machine learning) model. He referred me to the video series below:
Having all the videos in the series (Advice for Applying Machine Learning - What to Do Next 1-7) reviewed , I learned about the machine learning diagnostic issue. To Andries Ng's words, machine learning diagnostics is defined as: "a test that you can run to gain insight what is/isn't working with a learning algorithm and gain guidance as to how best to improve its (predictive) performance."
Machine learning diagnostics is built on the identification of "bias" and "variance" symptoms of predictive models. Based on the symptom's identification process, we can then define the prospective measures that will likely lead to better result.
So, what to do next then?
Based on the advice from lecture sets, the alternative measures are:
... and back to work, for sure :p
No comments:
Post a Comment