I have covered till softmax in course-2 of ML specialization.
I have been having this doubt as to how number of units are decided?
By a combination of experience (what has worked in the past with similar problems) and experimentation (knowing what to do when your first set of choices doesn’t work as well as you need them to). These topics are covered in DLS, but I’m not sure whether they are covered in MLS. In DLS, it is in Course 2 that Prof Ng really gives the picture of how to evaluate performance and then how to approach the problem when the performance is not good enough yet.
A natural progression would be to take DLS when you finish with MLS.
Thank you Mr.Paul…
I am now currently in week 2 of course 2 of MLS.
After completion of MLS, I have first planned to strengthen the concepts that I learnt in MLS by exposing myself to a lot of problem sets…
Could you please render me a resource which will help me achieve my goal?
The Kaggle website is a good place to look for ML problems. They have everything from relatively “intro” problems intended to get people up to speed and up to the state of the art challenges.
honestly completing first DLS and then exploring other websites like kaggle would be more beneficial to you. Kaggle coding is more tough to crack than doing assignment here as you are provided a metadata here. on kaggle you are just provided data and a sample submission and you need to code solutions from scratch, so make sure your SQL, python programming is strong when you explore kaggle.
What paul explained about DLS will help you strengthen understanding about data, architecture and other main features in neural network.
kaggle also uses other programming as well as statistical analyses.
1 Like